CN115061485A - Unmanned aerial vehicle guidance instruction generation method and system based on neural network - Google Patents
Unmanned aerial vehicle guidance instruction generation method and system based on neural network Download PDFInfo
- Publication number
- CN115061485A CN115061485A CN202210675669.5A CN202210675669A CN115061485A CN 115061485 A CN115061485 A CN 115061485A CN 202210675669 A CN202210675669 A CN 202210675669A CN 115061485 A CN115061485 A CN 115061485A
- Authority
- CN
- China
- Prior art keywords
- target
- unmanned aerial
- neural network
- aerial vehicle
- inclination angle
- 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
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 78
- 238000000034 method Methods 0.000 title claims abstract description 42
- 239000013598 vector Substances 0.000 claims abstract description 125
- 238000012549 training Methods 0.000 claims abstract description 60
- 238000003062 neural network model Methods 0.000 claims abstract description 50
- 239000002245 particle Substances 0.000 claims abstract description 48
- 238000005457 optimization Methods 0.000 claims abstract description 41
- 238000005096 rolling process Methods 0.000 claims abstract description 22
- 239000011159 matrix material Substances 0.000 claims description 30
- 230000014509 gene expression Effects 0.000 claims description 12
- 238000010606 normalization Methods 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 4
- 230000017105 transposition Effects 0.000 claims description 2
- 230000006870 function Effects 0.000 description 18
- 230000008569 process Effects 0.000 description 7
- 238000009795 derivation Methods 0.000 description 4
- 238000013461 design Methods 0.000 description 4
- 238000005070 sampling Methods 0.000 description 4
- 230000010354 integration Effects 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 239000000126 substance Substances 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000010485 coping Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012905 input function Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 239000011541 reaction mixture Substances 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D1/00—Control of position, course, altitude or attitude of land, water, air or space vehicles, e.g. using automatic pilots
- G05D1/08—Control of attitude, i.e. control of roll, pitch, or yaw
- G05D1/0808—Control of attitude, i.e. control of roll, pitch, or yaw specially adapted for aircraft
- G05D1/0816—Control of attitude, i.e. control of roll, pitch, or yaw specially adapted for aircraft to ensure stability
- G05D1/0825—Control of attitude, i.e. control of roll, pitch, or yaw specially adapted for aircraft to ensure stability using mathematical models
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D1/00—Control of position, course, altitude or attitude of land, water, air or space vehicles, e.g. using automatic pilots
- G05D1/10—Simultaneous control of position or course in three dimensions
- G05D1/101—Simultaneous control of position or course in three dimensions specially adapted for aircraft
- G05D1/106—Change initiated in response to external conditions, e.g. avoidance of elevated terrain or of no-fly zones
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Automation & Control Theory (AREA)
- Remote Sensing (AREA)
- Radar, Positioning & Navigation (AREA)
- Aviation & Aerospace Engineering (AREA)
- Algebra (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Mathematical Physics (AREA)
- Pure & Applied Mathematics (AREA)
- Control Of Position, Course, Altitude, Or Attitude Of Moving Bodies (AREA)
Abstract
The invention relates to a method and a system for generating an unmanned aerial vehicle guidance instruction based on a neural network, and relates to the technical field of unmanned aerial vehicle guidance. The method comprises the steps of determining an equation set according to a zero-effect miss distance under a rolling time domain optimization framework; obtaining a plurality of groups of target motion states according to a kinematic equation; obtaining an optimal inclination angle velocity vector, an optimal deflection angle velocity vector and a plurality of groups of aircraft motion states by adopting a particle swarm algorithm according to an equation set under each group of target motion states; training a neural network according to the motion state, the two optimal angular velocity vectors, the inclination angles and the deflection angular velocities at a plurality of moments to obtain an angular velocity neural network model and a velocity estimator; and obtaining a guidance instruction of the unmanned aerial vehicle at the current moment according to the motion state of the unmanned aerial vehicle at the current moment, the estimator and the neural network model. The method can obtain the guidance instruction of the unmanned aerial vehicle without the need of completely knowing the target motion information.
Description
Technical Field
The invention relates to the field of unmanned aerial vehicle guidance, in particular to an unmanned aerial vehicle guidance instruction generation method and system based on a neural network.
Background
The aircraft guidance law is a technical means for guiding the aircraft to approach the target and implementing interception according to state information such as the position, the speed and the like of the target, is one of key technologies of a fire control system for realizing accurate striking or interception guidance, and whether the design of the guidance law is reasonable plays a vital role in accurately intercepting the target or not.
Because of the advantages of simple form, high guidance precision and the like, the proportional guidance law is the most widely applied guidance law at present, but when the target has strong maneuverability, the guidance performance of the proportional guidance law is rapidly reduced, on the contrary, researchers have proposed various improvement measures, and novel guidance laws such as a sliding mode variable structure guidance law and a self-adaptive fuzzy guidance law are adopted, and meanwhile, the optimal guidance law can be more widely applied because the guidance problem with state constraint can be solved. The optimal guidance law is based on the optimal control theory, and according to the actual situation of aircraft guidance, a plurality of performance indexes including terminal miss distance, guidance time and energy consumption are considered, and the guidance law with optimal performance is obtained by optimizing a performance function. At present, the guidance law based on the linear quadratic optimization theory has entered into a practical stage, but since the optimal guidance law of global optimization requires that all target motion information are known, it is difficult to implement for the measurement elements of the aircraft, and since the terminal time needs to be estimated in real time in the guidance process, it is also very difficult under the condition that the target motion acceleration information is unknown, it is very important to research a guidance instruction generation method of the unmanned aircraft that does not need to know all target motion information.
Disclosure of Invention
The invention aims to provide a method and a system for generating a guidance instruction of an unmanned aerial vehicle based on a neural network, which do not need to obtain the guidance instruction of the unmanned aerial vehicle under the condition that target motion information is completely known.
In order to achieve the purpose, the invention provides the following scheme:
a method for generating unmanned aerial vehicle guidance instructions based on a neural network comprises the following steps:
determining an equation set according to the zero-effect miss distance under a rolling time domain optimization framework, wherein the equation set comprises a collaborative equation, a cross section conditional expression, an unmanned aerial vehicle guidance optimization model, an inclination angle speed local optimal guidance law equation and a deflection angle speed local optimal guidance law equation;
obtaining motion states of a plurality of groups of targets according to a target kinematics equation, wherein the motion states comprise an x coordinate, a y coordinate, a z coordinate, an inclination angle and a deflection angle;
obtaining an optimal inclination angle velocity vector, an optimal deflection angle velocity vector and motion states of a plurality of groups of unmanned aerial vehicles by adopting a particle swarm algorithm according to the equation set under the motion states of the targets;
training a neural network according to the motion states of the multiple groups of unmanned aerial vehicles, the motion states of the multiple groups of targets and the optimal inclination angle velocity vector by adopting an L-M algorithm to obtain an inclination angle velocity neural network model of the unmanned aerial vehicles;
training a neural network by adopting an L-M algorithm according to the motion states of the multiple groups of unmanned aerial vehicles, the motion states of the multiple groups of targets and the optimal deflection angle velocity vector to obtain an unmanned aerial vehicle deflection angle velocity neural network model;
training the neural network according to the inclination angles at multiple moments by adopting an L-M algorithm to obtain a target track inclination angle speed estimator;
training a neural network by adopting an L-M algorithm according to the deflection angle speeds at a plurality of moments to obtain a target track deflection angle speed estimator;
obtaining a state estimator of a target at a target moment according to the target track inclination angle and speed estimator, the target track deflection angle and speed estimator and the target kinematic equation;
obtaining a guidance instruction of the unmanned aerial vehicle at the current moment according to the motion state of the unmanned aerial vehicle at the current moment, the state estimator of the target moment, the inclination angle speed neural network model and the deflection angle speed neural network model; the guidance instructions include a pitch angular velocity instruction and a yaw angular velocity instruction.
An unmanned aerial vehicle guidance instruction generation system based on a neural network, comprising:
the system comprises an equation set determining module, a target miss distance determining module and a target miss distance determining module, wherein the equation set is determined according to a zero-effect miss distance under a rolling time domain optimization framework and comprises a collaborative equation, a cross section conditional expression, an unmanned aerial vehicle guidance optimization model, an inclination angle speed local optimal guidance law equation and a deflection angle speed local optimal guidance law equation;
the motion state initial module is used for obtaining motion states of a plurality of groups of targets according to a target kinematics equation, wherein the motion states comprise an x coordinate, a y coordinate, a z coordinate, an inclination angle and a deflection angle;
the optimal angular velocity vector determination module is used for obtaining an optimal angular velocity vector, an optimal deflection angular velocity vector and motion states of a plurality of groups of unmanned aerial vehicles by adopting a particle swarm algorithm according to the equation set under the motion states of all groups of targets;
the inclination angle velocity neural network model building module is used for training a neural network by adopting an L-M algorithm according to the motion states of the multiple groups of unmanned aerial vehicles, the multiple groups of target motion states and the optimal inclination angle velocity vector to obtain an inclination angle velocity neural network model;
the deflection angular velocity neural network model module is used for training a neural network by adopting an L-M algorithm according to the motion states of the multiple groups of unmanned aerial vehicles, the motion states of the multiple groups of targets and the optimal deflection angular velocity vector to obtain a deflection angular velocity neural network model;
the target track inclination angle speed estimator building module is used for training the neural network according to the inclination angles at multiple moments by adopting an L-M algorithm to obtain a target track inclination angle speed estimator;
the target track deflection angle speed estimator building module is used for training the neural network according to deflection angle speeds at a plurality of moments by adopting an L-M algorithm to obtain a target track deflection angle speed estimator;
the estimator determining module is used for obtaining a state estimator of the target at the target moment according to the target track inclination angle and speed estimator, the target track deflection angle and speed estimator and a target kinematic equation;
the guidance instruction determining module is used for obtaining a guidance instruction of the unmanned aerial vehicle at the current moment according to the motion state of the unmanned aerial vehicle at the current moment, the state estimator of the target at the target moment, the inclination angle and speed neural network model and the deflection angle and speed neural network model; the guidance instructions include a pitch angular velocity instruction and a yaw angular velocity instruction.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects: determining an equation set according to the zero-effect miss distance under a rolling time domain optimization framework; obtaining the motion states of a plurality of groups of targets according to a target kinematics equation; obtaining an optimal inclination angle velocity vector, an optimal deflection angle velocity vector and motion states of multiple groups of unmanned aerial vehicles by adopting a particle swarm algorithm according to an equation set under the motion states of all groups of targets; training a neural network by adopting an L-M algorithm according to the motion states of a plurality of groups of unmanned aerial vehicles, the motion states of a plurality of groups of targets and the optimal inclination angle velocity vector and the optimal deflection angle velocity vector to obtain an unmanned aerial vehicle inclination angle velocity neural network model and a deflection angle velocity neural network model; training a neural network by adopting an L-M algorithm according to the inclination angles and the deflection angles at a plurality of moments to obtain a target track inclination angle and speed estimator and a target track deflection angle and speed estimator; obtaining a state estimator of a target at a target moment according to the target track inclination angle and speed estimator, the target track deflection angle and speed estimator and a target kinematic equation; the method comprises the steps of obtaining a guidance instruction of the unmanned aerial vehicle at the current moment according to the motion state of the unmanned aerial vehicle at the current moment, the state estimator of a target at the target moment, an inclination angle speed neural network model and a deflection angle speed neural network model, designing a local optimal guidance law equation of an inclination angle speed and a local optimal guidance law equation of a deflection angle speed based on the neural network under a rolling time domain optimization framework, obtaining the guidance instruction of the unmanned aerial vehicle without the need of knowing all target motion information, and improving the generation speed of the guidance instruction.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a schematic three-dimensional motion of an unmanned aerial vehicle and a target;
fig. 2 is a flowchart of a method for generating an unmanned aerial vehicle guidance instruction based on a neural network according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
The embodiment of the invention provides a method for generating an unmanned aerial vehicle guidance instruction based on a neural network, which comprises the following steps as shown in figure 2:
step 101: and determining an equation set according to the zero-effect miss distance under a rolling time domain optimization framework. The equation set comprises a collaborative equation, a cross section conditional expression, an unmanned aerial vehicle guidance optimization model, an inclination angle speed local optimal guidance law equation and a deflection angle speed local optimal guidance law equation.
Step 102: and obtaining the motion states of the multiple groups of targets according to the target kinematics equation. The motion states include an x-coordinate, a y-coordinate, a z-coordinate, an angle of inclination, and an angle of deflection.
Step 103: and obtaining an optimal inclination angle velocity vector, an optimal deflection angle velocity vector and motion states of multiple groups of unmanned aerial vehicles by adopting a particle swarm algorithm according to the equation set under the motion states of the targets.
Step 104: and training a neural network by adopting an L-M algorithm according to the motion states of the multiple groups of unmanned aerial vehicles, the multiple groups of target motion states and the optimal inclination angle velocity vector to obtain an inclination angle velocity neural network model. Specifically, the motion states of a plurality of groups of unmanned aerial vehicles and the motion states of a plurality of groups of targets are used as input, the inclination angle and velocity vector of the unmanned aerial vehicle is used as output, and the first loss function value is minimum to be used as a target to train the neural network; the first loss function value is determined from the optimal tilt angular velocity vector and the tilt angular velocity vector,
step 105: and training a neural network by adopting an L-M algorithm according to the motion states of the multiple groups of unmanned aerial vehicles, the motion states of the multiple groups of targets and the optimal deflection angle velocity vector to obtain a deflection angle velocity neural network model. Specifically, the neural network is trained by taking the motion states of a plurality of groups of unmanned aerial vehicles and the motion states of a plurality of groups of targets as input, taking the deflection angle velocity vector of the unmanned aerial vehicle as output and taking the minimum second loss function value as a target; the second loss value is determined according to the yaw angular velocity vector and the optimal yaw angular velocity vector;
step 106: and training the neural network according to the inclination angles at multiple moments by adopting an L-M algorithm to obtain a target track inclination angle speed estimator. Specifically, the target track inclination angle and speed estimator is obtained by training a neural network by taking the first n-1 column of an inclination angle and speed matrix as input and the nth column of the inclination angle and speed matrix as output; the tilt angular velocity matrix comprises tilt angular velocities at a plurality of time instants.
Step 107: and training the neural network by adopting an L-M algorithm according to the deflection angular speeds at a plurality of moments to obtain a target track deflection angular speed estimator. Specifically, the target track yaw angular velocity estimator is obtained by training a neural network by taking the first n-1 column of a yaw angular velocity matrix as an input and the nth column of the yaw angular velocity matrix as an output; the yaw rate matrix includes yaw rate at a plurality of time instants.
Step 108: and obtaining the state estimator of the target at the target moment according to the target track inclination angle and speed estimator, the target track deflection angle and speed estimator and a target kinematic equation, wherein the target moment is the moment obtained by adding a prediction period to the current moment.
Step 109: and obtaining a guidance instruction of the unmanned aerial vehicle at the current moment according to the motion state of the unmanned aerial vehicle at the current moment, the state estimator of the target moment, the inclination angular velocity neural network model and the deflection angular velocity neural network model. The guidance instructions include a pitch angular velocity instruction and a yaw angular velocity instruction.
In practical application, the determining an equation set according to the zero-effect miss distance under a rolling time domain optimization framework specifically includes:
and determining the zero-effect miss distance in the three-dimensional plane according to the zero-effect miss distance in the two-dimensional plane.
And determining a local terminal index equation and a local performance function according to the zero-effect miss distance in the three-dimensional plane.
And determining an unmanned aerial vehicle guidance optimization model according to the local performance function and the local terminal index equation.
And processing and determining a collaborative equation, a cross section condition equation, an inclination angle speed local optimal guidance law equation and a deflection angle speed local optimal guidance law equation for the unmanned aerial vehicle guidance optimization model according to the Pontryagin minimum principle.
In practical application, the obtaining of the optimal inclination angle velocity vector, the optimal deflection angle velocity vector and the motion states of multiple groups of unmanned aerial vehicles by adopting a particle swarm algorithm according to the equation set in the motion state of each group of targets specifically comprises:
and calculating the integral of the collaborative equation in time to obtain a time collaborative equation.
And determining a fitness function according to the cross-section conditional expression and the unmanned aerial vehicle guidance optimization model.
For the motion states of any group of targets, randomly initializing the speeds of a group of particles and the unmanned aerial vehicle to obtain initial particles and initial speeds corresponding to the motion states of the targets; the particles are integral variables of the time-covariate equation.
And updating the initial particles corresponding to the motion state of the target and the initial speed by adopting a particle swarm algorithm by taking the fitness function minimum as a target to obtain the optimal particles corresponding to the motion state of the target.
And obtaining the optimal inclination angle speed, the optimal deflection angle speed and the motion state of the unmanned aerial vehicle corresponding to the motion state of the target according to the particles corresponding to the motion state of the target, the local optimal guidance law equation of the inclination angle speed and the local optimal guidance law equation of the deflection angle speed. And the motion state of the unmanned aerial vehicle is obtained by forward integration from the starting point of the prediction time domain to the end point of the prediction time domain according to the optimal control quantity, the motion state of the unmanned aerial vehicle at the sampling moment and the state equation.
And determining the optimal tilt angular velocity corresponding to the motion states of all the groups of targets as the optimal tilt angular velocity vector.
And determining the optimal yaw angular velocity corresponding to the motion states of all the groups of targets as the optimal yaw angular velocity vector.
And determining the motion states of the unmanned aerial vehicles corresponding to the motion states of all the groups of targets as the motion states of multiple groups of unmanned aerial vehicles.
In practical application, the training of the neural network according to the motion states of the multiple groups of unmanned aerial vehicles, the multiple groups of target motion states and the optimal inclination angle velocity vector by using the L-M algorithm to obtain the inclination angle velocity neural network model specifically includes:
and under the current training iteration times, inputting the motion states of the multiple groups of unmanned aerial vehicles and the motion states of the multiple groups of targets into a neural network under the current training iteration times to obtain an inclination angle velocity vector.
And sequentially carrying out normalization and transposition operations on the optimal inclination angle velocity vector to obtain a transposed optimal inclination angle velocity vector.
Calculating an error of the tilt angular velocity vector and the transposed optimal tilt angular velocity vector.
A loss function value is calculated from the error.
And judging whether the loss function value is smaller than a set threshold value or not according to the loss function value to obtain a first judgment result.
And if the first judgment result is yes, determining that the neural network under the current training iteration number is the inclination angle velocity neural network model.
And if the first judgment result is negative, updating the neural network under the current training iteration times according to the error to obtain an updated neural network, and judging whether the preset iteration times are reached to obtain a second judgment result.
If the second judgment result is yes, determining that the updated neural network is the inclination angular velocity neural network model.
And if the second judgment result is negative, determining that the updated neural network is the neural network under the next training iteration number and entering the next iteration.
In practical application, inputting the motion states of the multiple groups of unmanned aerial vehicles and the motion states of the multiple groups of targets into a neural network under the current training iteration number to obtain an inclination angle velocity vector, which specifically comprises the following steps:
determining x coordinates of all groups of the unmanned aerial vehicles as a first x input vector; determining y coordinates of all groups of the unmanned aerial vehicles as a first y input vector; determining z coordinates of all groups of the unmanned aerial vehicle as a first z input vector; determining the inclination angles of all the groups of the unmanned aerial vehicles as a first inclination angle input vector; and determining the deflection angles of all the groups of the unmanned aerial vehicles as a first deflection angle input vector.
Determining x coordinates of all groups of the target as a second x input vector; determining y coordinates of all groups of the target as a second y input vector; determining the z coordinates of all the groups of the target as a second z input vector; determining the tilt angles of all groups of the target as a second tilt angle input vector; the yaw angles of all groups of the target are determined as a second yaw angle input vector.
Respectively normalizing the first x input vector, the first y input vector, the first z input vector, the first inclination angle input vector, the first deflection angle input vector, the second x input vector, the second y input vector, the second z input vector, the second inclination angle input vector and the second deflection angle input vector to obtain a normalized vector set.
Combining the normalized vector set into a state matrix and transposing the state matrix to obtain the model input matrix;
and inputting the model input matrix into the neural network under the current training iteration times to obtain an inclination angle velocity vector.
In practical application, the obtaining of the guidance instruction of the unmanned aerial vehicle at the current time according to the motion state of the unmanned aerial vehicle at the current time, the state estimator of the target at the target time, the inclination angle and velocity neural network model and the deflection angle and velocity neural network model specifically includes:
and respectively carrying out normalization processing on the motion state of the unmanned aerial vehicle at the current moment and the state estimator of the target moment to obtain the normalized motion state of the unmanned aerial vehicle at the current moment and the normalized state estimator of the target moment.
And determining the normalized motion state of the unmanned aerial vehicle at the current moment and the normalized state estimator of the target at the target moment as an instruction input matrix.
And respectively inputting the command input matrix into the inclination angle speed neural network model and the deflection angle speed neural network model to obtain a normalized inclination angle speed and a normalized deflection angle speed.
And respectively carrying out reverse normalization on the normalized inclination angle speed and the normalized deflection angle speed to obtain a guidance instruction of the unmanned aerial vehicle at the current moment.
In practical application, the obtaining of the state estimator of the target at the target moment according to the target track inclination angle and speed estimator, the target track deflection angle and speed estimator, and the target kinematic equation specifically includes:
and integrating the target track inclination angle speed estimator to obtain an estimated value of the target track inclination angle.
And integrating the target track deflection angle speed estimator to obtain an estimated value of a target track deflection angle.
And inputting the estimated value of the target track inclination angle and the estimated value of the target track deflection angle into the target kinematics equation to obtain an equation to be solved.
And integrating the equation to be solved from the current moment to the target moment to obtain the state estimator of the target at the target moment. Specifically, the estimation value of the target track deflection angle speed and the estimation value of the target track inclination angle speed are integrated to obtain a target track deflection angle estimation value and a target track inclination angle estimation value, the estimation value of the inclination angle and the estimation value of the deflection angle at the time of t-t + Tp are input into a target kinematics equation, and the forward integration is carried out to obtain the state estimator of the target at the target time.
Aiming at the method, the invention also provides a system for generating the unmanned aerial vehicle guidance instruction based on the neural network, which comprises the following steps:
and the equation set determining module is used for determining an equation set according to the zero-effect miss distance under a rolling time domain optimization framework, wherein the equation set comprises a collaborative equation, a cross section conditional expression, an unmanned aerial vehicle guidance optimization model, an inclination angle speed local optimal guidance law equation and a deflection angle speed local optimal guidance law equation.
The motion state initial module is used for obtaining motion states of a plurality of groups of targets according to a target kinematics equation, and the motion states comprise an x coordinate, a y coordinate, a z coordinate, an inclination angle and a deflection angle.
And the optimal angular velocity vector determination module is used for obtaining the optimal inclination angular velocity vector, the optimal deflection angular velocity vector and the motion states of the multiple groups of unmanned aerial vehicles by adopting a particle swarm algorithm according to the equation set under the motion states of the various groups of targets.
And the inclination angle velocity neural network model building module is used for training a neural network by adopting an L-M algorithm according to the motion states of the multiple groups of unmanned aerial vehicles, the multiple groups of target motion states and the optimal inclination angle velocity vector to obtain an inclination angle velocity neural network model.
And the deflection angular velocity neural network model module is used for training a neural network by adopting an L-M algorithm according to the motion states of the multiple groups of unmanned aerial vehicles, the motion states of the multiple groups of targets and the optimal deflection angular velocity vector to obtain a deflection angular velocity neural network model.
And the target track inclination angle speed estimator building module is used for training the neural network according to the inclination angles at a plurality of moments by adopting an L-M algorithm to obtain the target track inclination angle speed estimator.
And the target track deflection angle speed estimator building module is used for training the neural network according to deflection angle speeds at a plurality of moments by adopting an L-M algorithm to obtain the target track deflection angle speed estimator.
And the estimator determining module is used for obtaining the state estimator of the target at the target moment according to the target track inclination angle and speed estimator, the target track deflection angle and speed estimator and a target kinematic equation.
A guidance instruction determining module, configured to obtain a guidance instruction of the unmanned aerial vehicle at the current time according to the motion state of the unmanned aerial vehicle at the current time, the state estimator of the target at the target time, the inclination angle and velocity neural network model, and the deflection angle and velocity neural network model; the guidance instructions include a pitch angular velocity instruction and a yaw angular velocity instruction.
The embodiment of the invention provides a more specific unmanned aerial vehicle guidance instruction generation method, which comprises the following specific steps:
1. establishing a system state equation according to a kinematics model of the unmanned aerial vehicle
In the three-dimensional guidance problem, the kinematic equations of the unmanned aerial vehicle and the target are shown as formula (1) and formula (2).
In the formulae (1) and (2), x M 、x T ，y M 、y T ，z M 、z T Coordinates of the unmanned aerial vehicle and the target on x, y and z axes of an inertial coordinate system are respectively, the upper dotted points represent that a time derivative is obtained for the coordinates, and theta M 、θ T Respectively the track inclination angle, psi, of the unmanned aerial vehicle and the target M 、ψ T Track deflection angle, v, of unmanned aerial vehicle and target, respectively M 、v T The velocities of the unmanned aerial vehicle and the target, respectively. A schematic diagram of the three-dimensional motion of the unmanned aerial vehicle and the target is shown in fig. 1.
Setting the control quantity
a is a track inclination angle speed instruction, u is a track deflection angle speed instruction,
the track inclination angle speed of the unmanned aerial vehicle is obtained by carrying out time derivation on the track inclination angle of the unmanned aerial vehicle,
the track deflection angle speed of the unmanned aerial vehicle is obtained by performing time derivation on the track deflection angle of the unmanned aerial vehicle, x represents the state of a guidance system, and x represents the state of the guidance system 1 、x 2 、x 3 Are respectively the difference value of the coordinate of the unmanned aerial vehicle and the target in the x, y and z axes, x 4 、x 5 、x 6 Are the difference values of the components of the speed of the unmanned aerial vehicle and the target on the x axis, the y axis and the z axis respectively,
establishing a state equation of a guidance system
In the formula (4), the reaction mixture is,
indicating that the guidance system state is time-derived,
meaning that the derivation of the target tilt angle speed over time is performed,
representing the time derivation of the target yaw angular velocity, the five coefficient matrixes are respectively:
the global performance function for designing the guidance problem is shown in equation (5).
Wherein, t f For guidance termination time, x 1 (t f )、x 2 (t f )、x 3 (t f ) Respectively representing the difference values of the unmanned aerial vehicle and the target at the moment of termination of guidance in x, y and z axis coordinates, a (t) representing a track inclination angle speed command at the moment t, and u (t) representing a track deflection angle speed command at the moment t.
2. Unmanned aerial vehicle guidance optimization model established under rolling time domain optimization framework
In order to solve the problem that the optimal terminal time is difficult to determine in the global optimal guidance problem, an unmanned aerial vehicle guidance optimization model is constructed based on a rolling time domain optimization framework. And (4) rolling time domain optimization, namely, solving an open-loop optimization problem in a limited time domain on line to obtain the current control action, and rolling forward until the control is finished.
The following assumptions are introduced based on the current time t:
assume that 1: when time τ > T + T P While the target is not maneuvering, i.e. target tilt velocity
Angular deflection velocity
Are all 0.
Assume 2: when time τ > T + T P And stopping the unmanned aerial vehicle, namely controlling the controlled variables a (tau) and u (tau) to be 0.
Wherein T is P Is the prediction period.
Based on assumptions 1 and 2, the amount of zero-effect miss introduced on the two-dimensional plane is as follows:
in the equation (6), r represents a relative distance between the unmanned aerial vehicle and the target, q represents a line-of-sight angle between the unmanned aerial vehicle and the target, and the addition points represent time derivatives. According to the formula (6), the zero-effect miss distance in the three-dimensional plane is deduced and described by the set motion state to obtain
Accordingly, the local terminal index [ phi ], [ alpha ] is x (t+T P )]
And a local performance function J [ x (T + T) P )]Wherein the content of the first and second substances,
representing the amount of miss at the moment of interest, x 1 (t+Tp)、x 2 (t+Tp)、x 3 (t + Tp) represents the difference between the coordinates of the unmanned aerial vehicle at the target time and the coordinates of the target in the x, y and z axes, x 4 (t+Tp)、x 5 (t+Tp)、x 6 (t + Tp) represents the difference in the components of the velocities of the UAV and the target in the x, y, and z axes, respectively.
Considering the dynamics constraint of the unmanned aerial vehicle, establishing an unmanned aerial vehicle guidance optimization model in a finite time domain:
deriving J [ x (T + T) according to Pontryagin minimum value principle P )]Minimum requirement:
column-write Hamilton function according to equation (8)
H(x,a,u,λ,τ)＝a 2 (τ)+u 2 (τ)+λ T f(x,a,u,τ)(10)
Wherein λ (τ) is a Lagrangian multiplier vector function,
λ T (τ)＝[λ 1 (τ),λ 2 (τ),…,λ 6 (τ)]
obtaining: the cooperative equation is as follows:
the cross-section conditions are as follows:
wherein the content of the first and second substances,
from the Pontryagin minimum principle, in the optimal control [ a ] * (τ),u * (τ)]In the above, the Hamilton function takes a minimum value. The derivatives of H (x, a, u, λ, τ) with respect to a (τ), u (τ) and the second derivative are obtained as equation (13).
H (x, a, u, lambda, tau) is in the condition that the controlled variables a (tau), u (tau) are not constrained by the boundary
Takes a minimum value, and when a (tau), u (tau) are bound by a boundary and
or
In the case of (a), H (x, a, u, λ, τ) is in
And
a minimum value is taken. From this, it can be determined that the optimal control expression with lagrange multipliers (covariates) is:
3. method for acquiring network training sample data based on particle swarm optimization-rolling time domain optimization
In order to adapt the neural network to various maneuvering behaviors which the target may take, a plurality of groups of target inclination angle speed and deflection angle speed time sequences are generated based on random signals, and the motion states of the plurality of groups of targets are obtained according to the formula (2).
And aiming at the motion state of each group of targets, performing rolling solution on the open-loop optimization problem described in the formula (8) by adopting a particle swarm algorithm according to a collaborative equation formula (11) and a cross section conditional formula (12) to obtain a plurality of groups of optimal guidance instruction sequences and recording related data.
the specific steps for solving the open-loop optimization problem formula (8) at time t are described as follows:
step 1: the equation (11) is integrated with respect to time τ to obtain equation (18)
Integrating variable c 1 ，c 2 ，c 3 ，c 4 ，c 5 ，c 6 Coded as vector c ═ c 1 ,c 2 ,…,c 6 ] T The p-th particle in the k-th iteration can be represented as
Step 2: randomly generating a set of initial particles
And initial velocity
Wherein p is more than or equal to 1 and less than or equal to N, and N is the population scale.
Step 3: computing covariates according to equation (18)
(t≤τ≤t+T P ) To do so by
And an initial state x (T) from T to T + T according to the expressions (14), (15), (16), (17) and (4) P Forward integration to find the state vector
Control quantity
(t≤τ≤t+T P ) And calculating a cross-sectional conditional expression based on (12)
And calculating a performance function according to equation (8)
Step 4: defining an error vector as a function of the cross-sectional conditional expression (12)
The fitness function takes the form of the sum of the 2-norm of the error vector and the performance function, defined as
Step 5: and selecting the individual optimal particles and the global optimal particles according to the particle fitness, and entering the next iteration. Wherein the individual optimal particle is
j is not greater than the current iteration number; the global optimum particle is
If the current iteration number k is larger than the maximum iteration number or the global optimal particle fitness is smaller than a certain small positive value, stopping iteration and switching to step9, otherwise, executing steps 6-step 8;
step 6: updating particle positions
Velocity of renewed particles
In the formula (22), rand 1 、rand 2 Is a pseudo-random number in (0,1), a hyperparameter l r1 、l r2 The method is characterized in that the learning rate is used, the flight speed of the particles to the individual optimal particles and the global optimal particles is determined by the value of the learning rate, the algorithm is converged faster when the learning rate is larger, but the algorithm is easy to fall into the local optimal state, and the value range is generally [0,4 ]]。
For adapting inertial weights
Wherein the hyperparameter w max 、w min Means maximum, minimumThe inertial weight.
Step 7: repeating step 3-step 4 to calculate the covariates
Control quantity
And a state vector
(t≤τ≤t+T P ) And finding a performance function
And fitness function
Step 8: selecting the individual optimal particles and the global optimal particles according to the particle fitness, and entering the next iteration;
step 9: the iteration is terminated, the globally optimal particles are output, and the optimal control sequence a is calculated according to the equations (14) and (15) * (τ)，u * (τ)，(t≤τ≤t+T P )。
The data required to be recorded in the solving process comprises the following steps: unmanned aerial vehicle motion state x at starting point of each control cycle M (kT C )，y M (kT C )，z M (kT C )，θ M (kT C )，ψ M (kT C ) Predicting the target motion state (known a priori) x at the end of the cycle T (kT C +T P )，y T (kT C +T P )，z T (kT C +T P )，θ T (kT C +T P )，ψ T (kT C +T P ) And the optimal control quantity a (kT) obtained by solving through the particle swarm optimization C )，u(kT C ) (obtained by the step 9), wherein k is 0,1,2, … and kT C ＜t f ，t f The guidance termination time for each solving process.
Respectively forming the recorded data into an input vector x M ，y M ，z M ，θ M ，ψ M ，x T ，y T ，z T ，θ T ，ψ T And target vectors a, u, for x M ，y M ，z M ，θ M ，ψ M ，x T ，y T ，z T ，θ T ，ψ T A, u are normalized respectively by x M For example, the processing method is
Wherein x is Mj Representing a vector x M The ith component of (2), normalizing the normalized state vector x M ′，y M ′，z M ′，θ M ′，ψ M ′，x T ′，y T ′，z T ′，θ T ′，ψ T ' composition model input matrix S in Where the number of rows and columns of the model input matrix is 10 and is the number of samples, which is denoted as N s 。
S in ＝[x M ′,y M ′,z M ′,θ M ′,ψ M ′,x T ′,y T ′,z T ′,θ T ′,ψ T ′] T (24)
4 designing guidance instruction on-line generator based on deep neural network
4.1 construction of deep neural networks and off-line training
In order to determine the functional relationship between the optimal control variable and the motion state, a deep neural network DNN is constructed with the control variables a and u as outputs a And DNN u . Experiments prove that when the number of the hidden layers is 5, the network fitting degree is high when the activation function is a hyperbolic tangent function, and the network convergence speed is high and the performance is good when the Levenberg-Marquardt (L-M) algorithm is adopted for training. By DNN a For example, the network offline training procedure based on the L-M algorithm is described as follows.
Step 1: setting the number of neurons in the hidden layer as n hi And i is 1,2, …,5, network parameters are initialized randomly.
B 6 ＝[b 6 ]∈R 1×1 b 6ij ＝rand(0,1)
Step 2: defining matrix operations
Wherein Λ ═ λ ij ]Is an arbitrary dimension matrix.
Calculating each hidden layer output and network actual output
Wherein 1 is m×n Representing an m x n dimensional matrix of elements all 1, H i out Representing the output of the i-th hidden layer, N out Representing the network output.
Abstract equation (25) into a functional form:
N out ＝f a (S in |ω a ) (26)
wherein
And laying out the network parameters as vectors for vectorized network parameters.
Calculating an error vector
Calculating a loss function
Step 3: when L (E) is less than a certain preset value, the training is terminated, the network parameters are output, otherwise step 4-step 5 are executed.
Step 4: calculating the network output error E with respect to the network parameter ω according to equation (29) a The Jacobian matrix of (1);
updating the network parameters according to equation (30);
step 5: if the training times reach the preset value, terminating the training and outputting the network parameters, otherwise executing step 2-step 3.
To this end, the shape ofNetwork model f described by equation (26) a (S in |ω a ) In the same way, f can be obtained u (S in |ω u )。
4.2 design System interference estimator
Due to the actual guidance process, x T (t+T P )，y T (t+T P )，z T (t+T P )，θ T (t+T P )，ψ T (t+T P ) When the motion state is difficult to know in advance, a system interference estimator is designed based on a single hidden layer neural network, and the target track inclination angle speed in a finite time domain is subjected to
Track yaw angular velocity
τ＝t,…,t+T P - Δ t is estimated online and from the real-time detected moving state x of the object T (t)，y T (t)，z T (t)，θ T (t)，ψ T (t) x can be inferred from the equation (2) of the target kinematics T (t+T P )，y T (t+T P )，z T (t+T P )，θ T (t+T P )，ψ T (t+T P ) The estimate of (2), i.e. the state of motion of the object detected in real time, is used to calculate the estimate. Because the working principle of the estimator is similar to the network training process described in section 4.1, only the detailed schemes of sample data updating, neural network construction, estimator output and the like are introduced in this section.
To be provided with
For example, defining a matrix of m × n dimensional historical information
Initializing the historical information matrix, i.e. commands, before guidance begins
At sampling time t (t ≧ Δ t), updating is performed according to equations (31) to (32)
And
respectively representing the historical information matrixes at the time t and the time t-delta t,
to represent
The ith row and the jth column of (g),
representing the measured track pitch angular velocity at time t-at. The history information matrix at each sampling time is updated according to equations (31) to (33).
To pair
Normalization processing is carried out according to the following steps:
to obtain
To be provided with
The first n-1 columns of (A) constitute a model input matrix S containing m samples and n-1 features in 。
To be provided with
As the desired output S of the network out 。
A neural network is constructed by a single hidden layer and a hyperbolic tangent activation function, network training is carried out by adopting an L-M algorithm, and the training process can refer to step 1-step 5 in 4.1 sections to obtain a functional relation N out ＝g(S in |ζ)
The most recent set of sampled data (the tilt angular velocity over a period of time)
Form a column vector
According to formula (35) pair
Carrying out normalization treatment to obtain
Will be provided with
Input function N out ＝g(S in ζ), compute network output
To pair
Inverse normalization is carried out to obtain
Then
I.e. the estimated value of the target track inclination angle speed at the current moment,
updating the column vector according to equations (38) to (39)
The formula (35) to the formula (39) are repeated, so that the estimation quantity of the inclination angle speed can be obtained in sequence
Similarly, the target track yaw rate estimator may be designed as such.
4.3 design guidance instruction Generator
At the sampling time t, the motion state x of the unmanned aerial vehicle is measured in real time M (t)，y M (t)，z M (t)，θ M (t)，ψ M (t), target state estimator
Using neural networks DNN a And DNN u And designing a guidance instruction generator. The method comprises the following steps:
step1 for x M (t)，y M (t)，z M (t)，θ M (t)，ψ M (t)，
Normalization is performed with x M (t) for example, the treatment method is
Composing an input vector
s in ＝[x′ M ,y′ M ,z′ M ,θ′ M ,ψ′ M ,x′ T ,y′ T ,z′ T ,θ′ T ,ψ′ T ] T (41)
Step2 converting s in As DNN a And DNN u Respectively calculate
Step3 pairs
Inverse normalization process
Guidance instruction for target time
The invention has the following technical effects:
1. under a rolling time domain optimization framework, a zero-effect miss distance design local optimal guidance law is introduced, the method does not need to acquire all the later motion states of the target and calculate the optimal terminal time according to the motion states, and can better adapt to an actual system through prediction and feedback, so that the robustness of a guidance strategy is improved.
2. And randomly generating a large number of target motion states, and respectively performing rolling solution on each group of target motion states by using a particle swarm algorithm to obtain a guidance training data set. The data acquisition and optimization solving method improves the solving precision of the guidance instruction and improves the capability of the subsequently designed guidance method for coping with various target maneuvers.
3. And constructing a neural network, performing offline learning on a plurality of groups of guidance training data obtained by rolling solution of the particle swarm algorithm, training to generate an unmanned aerial vehicle guidance network model, and designing a guidance instruction online generator according to the model. The method does not need to solve the open-loop optimization problem in an iterative manner, saves the instruction generation time, and further improves the on-line solving efficiency of the guidance law.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the description of the method part.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.
Claims (8)
1. A method for generating unmanned aerial vehicle guidance instructions based on a neural network is characterized by comprising the following steps:
determining an equation set according to the zero-effect miss distance under a rolling time domain optimization framework, wherein the equation set comprises a collaborative equation, a cross section conditional expression, an unmanned aerial vehicle guidance optimization model, an inclination angle speed local optimal guidance law equation and a deflection angle speed local optimal guidance law equation;
obtaining motion states of a plurality of groups of targets according to a target kinematics equation, wherein the motion states comprise an x coordinate, a y coordinate, a z coordinate, an inclination angle and a deflection angle;
obtaining an optimal inclination angle velocity vector, an optimal deflection angle velocity vector and motion states of a plurality of groups of unmanned aerial vehicles by adopting a particle swarm algorithm according to the equation set under the motion states of the targets;
training a neural network by adopting an L-M algorithm according to the motion states of the multiple groups of unmanned aerial vehicles, the motion states of the multiple groups of targets and the optimal inclination angle velocity vector to obtain an inclination angle velocity neural network model;
training a neural network by adopting an L-M algorithm according to the motion states of the multiple groups of unmanned aerial vehicles, the motion states of the multiple groups of targets and the optimal deflection angle speed vector to obtain a deflection angle speed neural network model;
training the neural network according to the inclination angles at multiple moments by adopting an L-M algorithm to obtain a target track inclination angle speed estimator;
training a neural network by adopting an L-M algorithm according to the deflection angle speeds at a plurality of moments to obtain a target track deflection angle speed estimator;
obtaining a state estimator of a target at a target moment according to the target track inclination angle and speed estimator, the target track deflection angle and speed estimator and the target kinematic equation;
obtaining a guidance instruction of the unmanned aerial vehicle at the current moment according to the motion state of the unmanned aerial vehicle at the current moment, the state estimator of the target moment, the inclination angle and speed neural network model and the deflection angle and speed neural network model; the guidance instructions include a pitch angular velocity instruction and a yaw angular velocity instruction.
2. The method for generating the unmanned aerial vehicle guidance instruction based on the neural network as claimed in claim 1, wherein the determining the equation set according to the zero-effect miss distance under the rolling time domain optimization framework specifically comprises:
determining the zero-effect miss distance in the three-dimensional plane according to the zero-effect miss distance in the two-dimensional plane;
determining a local terminal index equation and a local performance function according to the zero-effect miss distance in the three-dimensional plane;
determining an unmanned aerial vehicle guidance optimization model according to the local performance function and the local terminal index equation;
and processing and determining a collaborative equation, a cross section condition equation, an inclination angle speed local optimal guidance law equation and a deflection angle speed local optimal guidance law equation for the unmanned aerial vehicle guidance optimization model according to the Pontryagin minimum principle.
3. The method for generating the unmanned aerial vehicle guidance instruction based on the neural network as claimed in claim 1, wherein the obtaining of the optimal inclination angle velocity vector, the optimal deflection angle velocity vector and the motion states of the plurality of groups of unmanned aerial vehicles by the particle swarm algorithm according to the equation set in the motion states of the respective groups of targets specifically comprises:
calculating the integral of the collaborative equation in time to obtain a time collaborative equation;
determining a fitness function according to the cross-section conditional expression and the unmanned aerial vehicle guidance optimization model;
for the motion states of any group of targets, randomly initializing the speeds of a group of particles and the unmanned aerial vehicle to obtain initial particles and initial speeds corresponding to the motion states of the targets; the particles are integral variables of a time collaborative equation;
updating initial particles corresponding to the motion state of the target and the initial speed by adopting a particle swarm algorithm by taking the fitness function minimum as a target to obtain optimal particles corresponding to the motion state of the target;
obtaining an optimal inclination angle speed, an optimal deflection angle speed and a motion state of the unmanned aerial vehicle corresponding to the motion state of the target according to the particles corresponding to the motion state of the target, the inclination angle speed local optimal guidance law equation and the deflection angle speed local optimal guidance law equation;
determining the optimal tilt angular velocity corresponding to the motion states of all the groups of targets as the optimal tilt angular velocity vector;
determining the optimal yaw angular velocity corresponding to the motion states of all the groups of targets as the optimal yaw angular velocity vector;
and determining the motion states of the unmanned aerial vehicles corresponding to the motion states of all the groups of targets as the motion states of multiple groups of unmanned aerial vehicles.
4. The method for generating the unmanned aerial vehicle guidance instruction based on the neural network as claimed in claim 1, wherein the training of the neural network by using the L-M algorithm according to the motion states of the plurality of sets of the unmanned aerial vehicle, the motion states of the plurality of sets of the targets, and the optimal inclination velocity vector to obtain the inclination velocity neural network model specifically comprises:
under the current training iteration times, inputting the motion states of the multiple groups of unmanned aerial vehicles and the motion states of the multiple groups of targets into a neural network under the current training iteration times to obtain an inclination angle velocity vector;
sequentially carrying out normalization and transposition operations on the optimal inclination angle velocity vector to obtain a transposed optimal inclination angle velocity vector;
calculating an error of the tilt angular velocity vector and the transposed optimal tilt angular velocity vector;
calculating a loss function value from the error;
judging whether the loss function value is smaller than a set threshold value or not according to the loss function value to obtain a first judgment result;
if the first judgment result is yes, determining that the neural network under the current training iteration number is the inclination angle velocity neural network model;
if the first judgment result is negative, updating the neural network under the current training iteration times according to the error to obtain an updated neural network, and judging whether the preset iteration times are reached to obtain a second judgment result;
if the second judgment result is yes, determining that the updated neural network is the inclination angular velocity neural network model;
and if the second judgment result is negative, determining that the updated neural network is the neural network under the next training iteration number and entering the next iteration.
5. The method according to claim 4, wherein the step of inputting the motion states of the plurality of sets of unmanned aerial vehicles and the motion states of the plurality of sets of targets into the neural network under the current training iteration number to obtain the tilt angular velocity vector comprises:
determining x coordinates of all groups of the unmanned aerial vehicles as a first x input vector; determining y coordinates of all groups of the unmanned aerial vehicles as a first y input vector; determining z coordinates of all groups of the unmanned aerial vehicle as a first z input vector; determining the inclination angles of all the groups of the unmanned aerial vehicles as a first inclination angle input vector; determining deflection angles of all groups of the unmanned aerial vehicle as a first deflection angle input vector;
determining x coordinates of all groups of the target as a second x input vector; determining y coordinates of all groups of the target as a second y input vector; determining the z coordinates of all the groups of the target as a second z input vector; determining the tilt angles of all groups of the target as a second tilt angle input vector; determining deflection angles of all groups of the target as a second deflection angle input vector;
respectively normalizing the first x input vector, the first y input vector, the first z input vector, the first inclination angle input vector, the first deflection angle input vector, the second x input vector, the second y input vector, the second z input vector, the second inclination angle input vector and the second deflection angle input vector to obtain a normalized vector set;
combining the normalized vector set into a state matrix and transposing the state matrix to obtain the model input matrix;
and inputting the model input matrix into the neural network under the current training iteration times to obtain an inclination angle velocity vector.
6. The method for generating the unmanned aerial vehicle guidance instruction based on the neural network as claimed in claim 1, wherein the obtaining of the guidance instruction of the unmanned aerial vehicle at the current time according to the motion state of the unmanned aerial vehicle at the current time, the state estimator of the target at the target time, the inclination angular velocity neural network model and the yaw angular velocity neural network model specifically comprises:
respectively normalizing the motion state of the unmanned aerial vehicle at the current moment and the state estimator of the target moment to obtain the normalized motion state of the unmanned aerial vehicle at the current moment and the normalized state estimator of the target moment;
determining the normalized motion state of the unmanned aerial vehicle at the current moment and the normalized state estimator of the target at the target moment as command input matrixes;
inputting the command input matrix into the inclination angle speed neural network model and the deflection angle speed neural network model respectively to obtain a normalized inclination angle speed and a normalized deflection angle speed;
and respectively carrying out reverse normalization on the normalized inclination angle speed and the normalized deflection angle speed to obtain a guidance instruction of the unmanned aerial vehicle at the current moment.
7. The method for generating the unmanned aerial vehicle guidance instruction based on the neural network as claimed in claim 1, wherein the obtaining of the state estimator of the target at the target moment according to the target track inclination angle and speed estimator, the target track deflection angle and speed estimator, and the target kinematic equation specifically comprises:
integrating the target track inclination angle speed estimator to obtain an estimated value of a target track inclination angle;
integrating the target track deflection angle speed estimator to obtain an estimated value of a target track deflection angle;
inputting the estimated value of the target track inclination angle and the estimated value of the target track deflection angle into the target kinematics equation to obtain an equation to be solved;
and integrating the equation to be solved from the current moment to the target moment to obtain the state estimator of the target at the target moment.
8. An unmanned aerial vehicle guidance instruction generation system based on a neural network, comprising:
the system comprises an equation set determining module, a target miss distance determining module and a target miss distance determining module, wherein the equation set is determined according to a zero-effect miss distance under a rolling time domain optimization framework and comprises a collaborative equation, a cross section conditional expression, an unmanned aerial vehicle guidance optimization model, an inclination angle speed local optimal guidance law equation and a deflection angle speed local optimal guidance law equation;
the motion state initial module is used for obtaining motion states of a plurality of groups of targets according to a target kinematics equation, wherein the motion states comprise an x coordinate, a y coordinate, a z coordinate, an inclination angle and a deflection angle;
the optimal angular velocity vector determination module is used for obtaining the optimal angular velocity vector, the optimal deflection angular velocity vector and the motion states of a plurality of groups of unmanned aerial vehicles by adopting a particle swarm algorithm according to the equation set under the motion states of all groups of targets;
the inclination angle velocity neural network model building module is used for training a neural network by adopting an L-M algorithm according to the motion states of the multiple groups of unmanned aerial vehicles, the multiple groups of target motion states and the optimal inclination angle velocity vector to obtain an inclination angle velocity neural network model;
the deflection angle speed neural network model module is used for training a neural network according to the motion states of the multiple groups of unmanned aerial vehicles, the motion states of the multiple groups of targets and the optimal deflection angle speed vector by adopting an L-M algorithm to obtain a deflection angle speed neural network model;
the target track inclination angle speed estimator building module is used for training the neural network according to the inclination angles at multiple moments by adopting an L-M algorithm to obtain a target track inclination angle speed estimator;
the target track deflection angular velocity estimator building module is used for training the neural network according to deflection angular velocities at multiple moments by adopting an L-M algorithm to obtain a target track deflection angular velocity estimator;
the estimator determining module is used for obtaining a state estimator of the target at the target moment according to the target track inclination angle and speed estimator, the target track deflection angle and speed estimator and the target kinematic equation;
a guidance instruction determining module, configured to obtain a guidance instruction of the unmanned aerial vehicle at the current time according to the motion state of the unmanned aerial vehicle at the current time, the state estimator of the target at the target time, the inclination angle and velocity neural network model, and the deflection angle and velocity neural network model; the guidance instructions include a pitch angular velocity instruction and a yaw angular velocity instruction.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210675669.5A CN115061485A (en) | 2022-06-15 | 2022-06-15 | Unmanned aerial vehicle guidance instruction generation method and system based on neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210675669.5A CN115061485A (en) | 2022-06-15 | 2022-06-15 | Unmanned aerial vehicle guidance instruction generation method and system based on neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115061485A true CN115061485A (en) | 2022-09-16 |
Family
ID=83201355
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210675669.5A Pending CN115061485A (en) | 2022-06-15 | 2022-06-15 | Unmanned aerial vehicle guidance instruction generation method and system based on neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115061485A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117390830A (en) * | 2023-09-25 | 2024-01-12 | 中国人民解放军海军航空大学 | Unmanned aerial vehicle cross-platform guidance simulation method, system and medium |
-
2022
- 2022-06-15 CN CN202210675669.5A patent/CN115061485A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117390830A (en) * | 2023-09-25 | 2024-01-12 | 中国人民解放军海军航空大学 | Unmanned aerial vehicle cross-platform guidance simulation method, system and medium |
| CN117390830B (en) * | 2023-09-25 | 2024-05-28 | 中国人民解放军海军航空大学 | Unmanned aerial vehicle cross-platform guidance simulation method, system and medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111752280B (en) | Multi-unmanned ship formation fixed time control method based on finite time uncertainty observer | |
| CN111240345B (en) | Underwater robot trajectory tracking method based on double BP network reinforcement learning framework | |
| CN109901403B (en) | Autonomous underwater robot neural network S-plane control method | |
| Wang et al. | Path planning under constraints and path following control of autonomous underwater vehicle with dynamical uncertainties and wave disturbances | |
| CN110928310A (en) | Unmanned ship navigation following fixed time formation control method | |
| CN109739090A (en) | A kind of autonomous type underwater robot neural network intensified learning control method | |
| CN112462792B (en) | Actor-Critic algorithm-based underwater robot motion control method | |
| CN113377121B (en) | Aircraft intelligent disturbance rejection control method based on deep reinforcement learning | |
| CN111240344B (en) | Autonomous underwater robot model-free control method based on reinforcement learning technology | |
| CN112947505B (en) | Multi-AUV formation distributed control method based on reinforcement learning algorithm and unknown disturbance observer | |
| CN111176122B (en) | Underwater robot parameter self-adaptive backstepping control method based on double BP neural network Q learning technology | |
| Song et al. | Guidance and control of autonomous surface underwater vehicles for target tracking in ocean environment by deep reinforcement learning | |
| CN111445498A (en) | Target tracking method adopting Bi-L STM neural network | |
| CN114675673B (en) | Method and system for tracking moving target in air | |
| CN115061485A (en) | Unmanned aerial vehicle guidance instruction generation method and system based on neural network | |
| CN106863297A (en) | A kind of accurate approach method of space rope system robot vision | |
| CN114004035B (en) | Target tracking control method for unmanned surface vehicle | |
| CN114019985B (en) | Unmanned rudder direction control design method based on fractional order PID and particle swarm algorithm | |
| CN115480580A (en) | NMPC-based underwater robot path tracking and obstacle avoidance control method | |
| CN114237254B (en) | Unmanned rudder PID control design method based on Smith predictor and particle swarm optimization | |
| CN116088311A (en) | Unmanned helicopter second-order uncertain sliding mode control method based on adaptive neural network extended state observer | |
| CN110703792B (en) | Underwater robot attitude control method based on reinforcement learning | |
| CN116774576A (en) | Underwater vehicle dynamics black box modeling method based on neural network indirect estimation | |
| CN114118371A (en) | Intelligent agent deep reinforcement learning method and computer readable medium | |
| Gong et al. | Three-dimensional optimal trajectory tracking control of underactuated AUVs with uncertain dynamics and input saturation |
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 |