CN108919828A - A kind of aerial vehicle trajectory optimization method based on artificial memory - Google Patents
A kind of aerial vehicle trajectory optimization method based on artificial memory Download PDFInfo
- Publication number
- CN108919828A CN108919828A CN201810768925.9A CN201810768925A CN108919828A CN 108919828 A CN108919828 A CN 108919828A CN 201810768925 A CN201810768925 A CN 201810768925A CN 108919828 A CN108919828 A CN 108919828A
- Authority
- CN
- China
- Prior art keywords
- memory
- trial solution
- time
- formula
- aerial vehicle
- 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
Classifications
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05D—SYSTEMS FOR CONTROLLING OR REGULATING NON-ELECTRIC VARIABLES
- G05D1/00—Control of position, course or altitude of land, water, air, or space vehicles, e.g. automatic pilot
- 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
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B13/00—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion
- G05B13/02—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric
- G05B13/04—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators
- G05B13/042—Adaptive control systems, i.e. systems automatically adjusting themselves to have a performance which is optimum according to some preassigned criterion electric involving the use of models or simulators in which a parameter or coefficient is automatically adjusted to optimise the performance
Abstract
The present invention proposes a kind of aerial vehicle trajectory optimization method based on artificial memory, belongs to the field of aerial vehicle trajectory design.The present invention has studied artificial memory's track optimizing method to solve the problems, such as that existing aerial vehicle trajectory optimization method convergence precision is low, convergence rate is slow.It models traditional track optimizing problem by memory principle；Guarantee the randomness of trial solution in each iteration using trial solution state evolution operator；Resettle stimulus type judgment models and memory more new model；Finally aerial vehicle trajectory optimization problem is studied using artificial memory's optimization method.The present invention is used for the aerial vehicle trajectory optimization design under the conditions of Complex Constraints.
Description
Technical field
The present invention relates to a kind of aerial vehicle trajectory optimization method based on artificial memory, belongs to aerial vehicle trajectory optimization design
Technical field.
Background technique
Motors in boost phase penetration Optimal design of trajectory is to solve for the nonlinear optimal control problem comprising path and end conswtraint.For rail
There are many method of mark optimization, are broadly divided into analytic method and numerical method two major classes, and wherein numerical method is sent out from computer technology height
A kind of method booming therewith after exhibition.Stryk, Betts et al. summarize and analyze the numerical method of Today, most, two people
Numerical method is divided into two class of indirect method and direct method.Indirect method is based on not to performance index function direct searching optimization
Optimal control problem is converted Hamilton boundary value problem by Pontryagin maximal principle.Direct method uses parametric method
Non-Linear Programming (Nonlinear programming, NLP) is converted by the optimal control problem solution of continuous space to ask
Topic, obtains optimal trajectory by numerical solution nonlinear programming problem.Direct method is more more extensive than the application of indirect method, commonly uses
It includes method etc. that method, which has direct shooting method, multiple direct shooting method, point collocation, differential,.Harmony grace rice, Chen Cong, yellow Guoqiang etc. are also right
Track optimizing method has carried out deep analysis, has carried out standardization description to track optimizing problem, has been set forth different
Track optimizing method classification, summarizes the latest Progress in track optimizing field in recent years.
It is generally believed that numerical method can be divided into indirect method and direct method according to the method for transformation of problem, according to whether relying on ladder
Degree information again can approximation be divided into two major classes：Precise search algorithm and random search algorithm.
So-called precise search algorithm refers to the searching method based on gradient information.Iterative search of such algorithm in optimizing
In the process, the best gradient side at current time is determined by solving the gradient of the penalty function based on performance indicator and constraint construction
To being continued searching at the iteration moment of next step according to the direction, until find optimal solution.It should be noted that precise search
For algorithm due to carrying out optimizing using gradient information, optimal solution is usually locally optimal solution, not can guarantee Global Optimality.
More and more there is multivariable, multiple target, multiple constraint, non-linear, multiple extremum, excellent in aerial vehicle trajectory optimization design
Change target and the irregular complicated optimum problem of constraint type, make traditional optimization algorithm --- precise search algorithm faces sternness and chooses
War.In recent years, with the continuous development of computer technology, researchers to genetic algorithm, ant group algorithm, particle swarm algorithm,
The stochastic search methods such as differential evolution algorithm extensively and profoundly study.
Summary of the invention
The invention aims to solve the problems, such as that existing track optimizing method Global Optimality is lower, a kind of base is provided
In the aerial vehicle trajectory optimization method of artificial memory.
The purpose of the present invention is achieved through the following technical solutions：A kind of aerial vehicle trajectory optimization side based on artificial memory
Method, it includes the following steps：
Step 1：Establish the track optimizing problem model of standard；
Step 2：Trial solution state evolution operator is designed, provides a large amount of trial solution for each iterative search；
Step 3：The target function value obtained using trial solution changes the type to judge stimulation；
Step 4：Design forgets and memory more new model；
Step 5：Track optimizing is carried out to aircraft using artificial memory's algorithm.
Further, the step 1 is specially：
The track optimizing problem model of the standard is：
Wherein, X=(x1,x2,…,xn) it is optimal control vector, S is feasible zone or search space, RnIndicate n dimension
The space Euclidean, f (X) are objective function, gi(X) >=0 inequality constraints is indicated, K is the number of inequality constraints；
The objective function of above-mentioned optimization problem is rewritten as：
In formula, FmaxIt is the positive number to differ greatly with f (X) magnitude, is used for screening and is unable to satisfy the constrained exploration of institute
Solution, consequently facilitating system carries out rejecting or other processing to it；
Memory elements are designed to data burst structure, including trial solution, memory residue, system time, memory state and something lost
Forget state, i.e.,
Mi=(Xi,mi,ti,si,fi), i=1,2 ..., N
In formula, point XiIndicate trial solution, N is memory elements number；miTo remember residue, note of the characterization memory elements to stimulation
Recall degree, mi≥0；tiFor system time, that is, search for the time carried out；siFor memory state, there are I, tri- states of S, L, difference
For it is instantaneous, in short-term with long-term memory state；fiTo forget state, work as fiShow memory elements M when=1iTrial solution X is forgotteni, work as fi
It then indicates clearly to remember when=0.
Further, the step 2 is specially：
Assuming that during the optimal solution search of the optimization problem, at current time (t), point XiProperty will receive it
The influence of its dotted state, enables the set that other points are constituted when previous moment (t-1), i.e. trial solution collection is combined into
In order to guarantee global convergence and convergence rate, the trial solution search strategy is as follows：
1) Average Strategy
A certain number of memory elements are possessed in long-term memory system at this time, i.e., good trial solution；From wherein random
L trial solution is picked out, enables the trial solution at current time (t) for the average value of L trial solution in previous moment (t-1), i.e. point
XiCorresponding trial solution value is：
In formula,i1,i2,…,iLFor from trial solution set X
The serial number of L trial solution of middle taking-up, n are point XiIn variable number；
2) orthogonal strategy
Two memory elements, serial number i are randomly selected from trial solution set X at this time1With i2, corresponding trial solution carries out orthogonal
Weighting operations, then point XiCorresponding trial solution value is：
In formula, a is the random number between 0~1, i.e. a=rand (0,1)；
3) projection strategy
Two memory elements, serial number i are randomly selected from trial solution set X at this time1With i2, correspond to trial solution and projected
It operates, then point XiCorresponding trial solution value is：
4) expanding policy
A memory elements, serial number k are randomly selected from trial solution set X at this time, corresponding trial solution is extended operation,
Then point XiCorresponding trial solution value is：
In formula, coefficient r1=rand (- 1,1)；
5) Chevy strategy
Choose the optimal trial solution of previous moment and the optimal trial solution at current time respectively from trial solution set X at this timeCorresponding trial solution carries out Chevy operation, then point XiCorresponding trial solution value is：
In formula, coefficient r2=rand (0.4,0.6), r3=1.3rand (0,1), r4=1.8rand (0,1)；
Above-mentioned 5 strategies can be all executed randomly in each search, each after entire Optimizing Search process
The number that strategy is performed is of substantially equal.
Further, the step 3 is specially：
1) effective stimulus
For point XiIf compared with previous moment (t-1), target function value reduces in the search at current time (t)
, i.e.,
Then illustrate the trial solution quality at current time better than previous moment, which is effective stimulus；
2) ineffective stimulation
Correspondingly, for point XiIf in the search at current time (t), compared with previous moment (t-1), objective function
Value does not reduce, i.e.,
Then illustrate that the trial solution quality at current time has no improvement, which is ineffective stimulation.
Further, the step 4 is specially：
Forgetting modelling is carried out using Ebbinghaus forgetting curve model, then forgetting function is：
In formula, t is the time, since 0；Δ t is taken as regular time increment；α > 0 is to forget velocity coeffficient；
Known effective stimulus can be improved memory, and memory residue is enabled to increase, then enabling the increment of memory residue is Δ mi
(t)；And the absolute value of target function value incrementBigger, then the stimulation before showing the stimulation ratio is more effective, intuitively
It is presented as that stimulation leads to the increment Delta m for remembering residuei(t) bigger；Then, increased using the memory residue of proportional relationship
Measure Δ mi(t) it is：
In formula, positive number h is effective stimulus proportionality coefficient, is taken as constant, is become for adjusting memory residue caused by stimulation
Change size；
Remember at this time and has just been divided into two parts：The new memory increment that existing old memory and stimulation generate, then remembers residual
Value is stayed to be expressed as
mi(t+ Δ t)=mi(t)+Δmi(t)
In addition, memory should also include the memory part to old memory, according to Ebbinghaus forgetting curve it is found that memory
Forgetting law be it is fixed, the speed only forgotten is different；
In order to distinguish old memory and to the memory of old memory, the memory residue at current time is expressed as form：
mi(t+ Δ t)=mi(t)e-αΔt+λmi(t)e-bΔt
In formula, function lambda mi(t)e-bΔtFor the memory value of memory, b is constant and b > α, that is, the forgetting speed recalled is faster；
λ is a positive coefficient, for adjusting the forgetting speed recalled, enables λ < 1, that is, recalling can not be more than the memory of original memory
Residue；
Then above formula can be rewritten as
mi(t+ Δ t)=mi(t)(e-αΔt+λe-bΔt)
The links for comprehensively considering memory models, finally obtain the more new model of total recall, and expression formula is
mi(t+ Δ t)=mi(t)(e-αΔt+λe-bΔt)+Δmi(t)
Enable β=e-αΔt+λe-bΔt, when time increment Delta t is fixed value, then β is also fixed constant, and then memory updates
Model simplification is following form：
In formula, factor betaI,βS,βLIt is respectively instantaneous, in short-term with the forgetting speed of long-term memory, by three memory systems
The forgetting speed of memory is it is found that 0 < βL< βS< βI< 1.
Further, the step 5 is specially：
Wherein the setting of parameters is as follows：
1) incremental time Δ t=0.1, system sweeps number or maximum number of iterations is G=10000；
2) element number of optimal control vector X, i.e. n=100；
3) memory elements number N=100, the memory elements number L=3 selected in long-term memory system；
4) effective stimulus proportionality coefficient h=1.2；
5) instantaneous, in short-term with the forgetting velocity coeffficient β of long-term memoryI=0.55, βS=0.35, βL=0.15.
Advantages of the present invention：The present invention devises a kind of aerial vehicle trajectory optimization method based on artificial memory.The invention
The track optimizing problem model for establishing standard, devise be made of immediate memory, short-term memory, long-term memory system it is artificial
Memory system forgets the more new model of model foundation memory using great this of Chinese mugwort guest, finally uses trial solution state evolution operator pair
The trial solution of each iteration is designed, and is obtained the trial solution with stronger randomness, is finally obtained track optimizing problem
Globally optimal solution.The present invention studies aerial vehicle trajectory optimization problem using the artificial memory based on memory principle, effectively
The Global Optimality for improving track optimizing has broad application prospects in the Optimal design of trajectory field comprising Complex Constraints.
Detailed description of the invention
Fig. 1 is the flow chart of the aerial vehicle trajectory optimization method of the present invention based on artificial memory.
Specific embodiment
Technical solution in the embodiment of the present invention that following will be combined with the drawings in the embodiments of the present invention carries out clear, complete
Ground description, it is clear that described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Based on this
Embodiment in invention, every other reality obtained by those of ordinary skill in the art without making creative efforts
Example is applied, shall fall within the protection scope of the present invention.
In conjunction with Fig. 1, the present invention proposes a kind of aerial vehicle trajectory optimization method based on artificial memory, it includes following step
Suddenly：
Step 1：The track optimizing problem model of standard is established, wherein objective function takes the form of penalty function；
Step 2：Trial solution state evolution operator is designed, provides a large amount of trial solution for each iterative search, and provide 5
Kind search strategy, guarantees the randomness of search；
Step 3：The target function value obtained using trial solution changes the type to judge stimulation；
Step 4：The forgetting model of memory is designed according to Ebbinghaus forgetting curve model, then derives note according to this
The more new model recalled；
Step 5：Track optimizing is carried out to aircraft using artificial memory's algorithm.
Track optimizing problem model is in step 1：
Wherein, X=(x1,x2,…,xn) it is optimal control vector, S is feasible zone or search space, RnIndicate n dimension
The space Euclidean, f (X) are objective function, gi(X) >=0 inequality constraints is indicated, K is the number of inequality constraints.
In order to facilitate use, the objective function of above-mentioned optimization problem is rewritten as
In formula, FmaxBe one and differ biggish positive number with f (X) magnitude, be used for screen be unable to satisfy institute it is constrained that
A little trial solutions, consequently facilitating system carries out rejecting or other processing to it.
Memory elements are designed to data burst structure, including trial solution, memory residue, system time, memory state and something lost
Forget state, i.e.,
Mi=(Xi,mi,ti,si,fi), i=1,2 ..., N
In formula, point XiIndicate trial solution, N is memory elements number；miTo remember residue, note of the characterization memory elements to stimulation
Recall degree, mi≥0；tiFor system time, that is, search for the time carried out；siFor memory state, there are I, tri- states of S, L, difference
For it is instantaneous, in short-term with long-term memory state；fiTo forget state, work as fiShow memory elements M when=1iTrial solution X is forgotteni, work as fi
Then indicate that there are also more clearly remember when=0.
In step 2, the design process of trial solution state evolution operator is as follows：
Assuming that during the optimal solution search of above-mentioned optimization problem, at current time (t), point XiProperty will receive it
The influence of its dotted state.The set that other points are constituted when previous moment (t-1) is enabled, i.e. trial solution collection is combined into
In order to guarantee global convergence and convergence rate, trial solution search strategy given herein is as follows.
1) Average Strategy
A certain number of memory elements are possessed in long-term memory system at this time, i.e., good trial solution.From wherein random
L trial solution is picked out, enables the trial solution at current time (t) for the average value of L trial solution in previous moment (t-1), i.e. point
XiCorresponding trial solution value is：
In formula,i1,i2,…,iLFor from trial solution set X
The serial number of L trial solution of middle taking-up, n are point XiIn variable number.
2) orthogonal strategy
Two memory elements, serial number i are randomly selected from trial solution set X at this time1With i2, corresponding trial solution carries out orthogonal
Weighting operations, then point XiCorresponding trial solution value is：
In formula, a is the random number between 0~1, i.e. a=rand (0,1).
3) projection strategy
Two memory elements, serial number i are randomly selected from trial solution set X at this time1With i2, correspond to trial solution and projected
It operates, then point XiCorresponding trial solution value is：
4) expanding policy
A memory elements, serial number k are randomly selected from trial solution set X at this time, corresponding trial solution is extended operation,
Then point XiCorresponding trial solution value is：
In formula, coefficient r1=rand (- 1,1).
5) Chevy strategy
Choose the optimal trial solution of previous moment and the optimal trial solution at current time respectively from trial solution set X at this timeCorresponding trial solution carries out Chevy operation, then point XiCorresponding trial solution value is：
In formula, coefficient r2=rand (0.4,0.6), r3=1.3rand (0,1), r4=1.8rand (0,1).
Above-mentioned 5 strategies can be all executed randomly in each search, each after entire Optimizing Search process
The number that strategy is performed is of substantially equal.Each strategy emphasizes particularly on different fields, and Average Strategy, which is to make full use of, has optimal trial solution,
It calculates simply, but is easily trapped into local optimum.Orthogonal strategy, projection strategy and expanding policy randomness are strong, can be to avoid part
Optimal problem, but not can guarantee convergence rate.Chevy strategy performance compromise, but calculation amount is larger.
In summary analysis can generate N number of exploration using 5 search strategies it is found that souning out Solution operator before search starts every time
Solution, has on the one hand fully ensured that convergent rapidity, has on the other hand ensured the randomness of search, can finally enable search smooth
It carries out and quickly obtains optimal solution.
In step 3, the stimulus type judgment method that trial solution generates is as follows：
As previously mentioned, stimulation (situation of change of target function value) that algorithm is generated by trial solution judges its type.
1) effective stimulus
For point XiIf compared with previous moment (t-1), target function value reduces in the search at current time (t)
, i.e.,
Then illustrate the trial solution quality at current time better than previous moment, which is effective stimulus.
2) ineffective stimulation
Correspondingly, for point XiIf in the search at current time (t), compared with previous moment (t-1), objective function
Value does not reduce, i.e.,
Then illustrate that the trial solution quality at current time has no improvement, which is ineffective stimulation.
In step 4, the update model inference process of memory is as follows：
German psychologist H.Ebbinghaus proposes famous Ebbinghaus and forgets song according to years of researches
Line, after which can describe brain by outside stimulus, memory decays the process of (forgetting) as time goes by and gradually.
The present invention carries out forgetting modelling using the model, and then forgetting function is：
In formula, t is the time, since 0；Δ t is taken as regular time increment；α > 0 is to forget velocity coeffficient.It needs
Bright, for the same stimulation, in TM, STM and LTM system, the speed that the memory generated passes into silence is not identical, because
This, which forgets velocity coeffficient, 3 kinds, respectively αI,αS,αL, and αI> αS> αL, i.e., remember in TM, STM and LTM system and pass into silence
Speed show slower and slower trend.
According to memory models it is found that on the one hand memory can gradually decay with the time, if had during another aspect
Effect stimulation can also enhance memory.As shown in the above, memory can be improved in effective stimulus, enables memory residue increase, enables
The increment for remembering residue is Δ mi(t).And the absolute value of target function value incrementIt is bigger, then show the stimulation ratio
Stimulation before is more effective, is intuitively presented as that stimulation leads to the increment Delta m for remembering residuei(t) bigger.Then, using just
The memory residue increment Delta m of proportionate relationshipi(t) it can be written as：
In formula, positive number h is effective stimulus proportionality coefficient, is taken as constant, is become for adjusting memory residue caused by stimulation
Change size.
Remember at this time and has just been divided into two parts：The new memory increment that existing old memory and stimulation generate, then remembers residual
Value is stayed to be expressed as
mi(t+ Δ t)=mi(t)+Δmi(t)
In addition, memory should also include the memory part to old memory.According to Ebbinghaus forgetting curve it is found that memory
Forgetting law be it is fixed, the speed only forgotten is different.
In order to distinguish old memory and to the memory of old memory, the memory residue at current time is expressed as form：
mi(t+ Δ t)=mi(t)e-αΔt+λmi(t)e-bΔt
In formula, function lambda mi(t)e-bΔtFor the memory value of memory, b is constant and b > α, that is, the forgetting speed recalled is faster；
λ is a positive coefficient, for adjusting the forgetting speed recalled, enables λ < 1, that is, recalling can not be more than the memory of original memory
Residue.
Then above formula can be rewritten as
mi(t+ Δ t)=mi(t)(e-αΔt+λe-bΔt)
The links of memory models are comprehensively considered, such as have forgotten, recall, memory-enhancing effect (by effective stimulus), most
The more new model of total recall is obtained eventually, and expression formula is
mi(t+ Δ t)=mi(t+ Δ t)=mi(t)(e-αΔt+λe-bΔt)+Δmi(t)
Enable β=e-αΔt+λe-bΔt, when time increment Delta t is fixed value, then β is also fixed constant, and then memory updates
Model simplification is following form：
In formula, factor betaI,βS,βLIt is respectively instantaneous, in short-term with the forgetting speed of long-term memory, by three memory systems
The forgetting speed of memory is it is found that 0 < βL< βS< βI< 1.
In step 5, artificial memory's algorithm is applied in aerial vehicle trajectory optimization problem.The wherein setting of parameters
As follows：
1) incremental time Δ t=0.1, system sweeps number or maximum number of iterations is G=10000.
2) element number of dominant vector X, i.e. n=100.
3) memory elements number N=100, the memory elements number L=3 selected in long-term memory system.
4) effective stimulus proportionality coefficient h=1.2.
5) instantaneous, in short-term with the forgetting velocity coeffficient β of long-term memoryI=0.55, βS=0.35, βL=0.15.
The present invention is proposed for the deficiency that existing track optimizing method Global Optimality is difficult to ensure, by artificial memory's original
It ought to be used in the track optimizing problem model amount of standard, traditional track optimizing problem be studied, by using difference
Trial solution search strategy, ensure that the randomness of algorithm, effectively increase the convergence rate of track optimizing, and further decrease
There is the probability of local minimum phenomenon.Meanwhile trial solution state evolution operator effectively improves the complete of track optimizing problem
Office's optimality, it is ensured that converge within given precision, the practical application for artificial Memory algorithm provides support.
Above to a kind of aerial vehicle trajectory optimization method based on artificial memory provided by the present invention, detailed Jie has been carried out
It continues, used herein a specific example illustrates the principle and implementation of the invention, and the explanation of above embodiments is only
It is to be used to help understand method and its core concept of the invention；At the same time, for those skilled in the art, according to this hair
Bright thought, there will be changes in the specific implementation manner and application range, in conclusion the content of the present specification should not manage
Solution is limitation of the present invention.
Claims (6)
1. a kind of aerial vehicle trajectory optimization method based on artificial memory, which is characterized in that it includes the following steps：
Step 1：Establish the track optimizing problem model of standard；
Step 2：Trial solution state evolution operator is designed, provides a large amount of trial solution for each iterative search；
Step 3：The target function value obtained using trial solution changes the type to judge stimulation；
Step 4：Design forgets and memory more new model；
Step 5：Track optimizing is carried out to aircraft using artificial memory's algorithm.
2. the aerial vehicle trajectory optimization method according to claim 1 based on artificial memory, it is characterised in that：The step
One is specially：
The track optimizing problem model of the standard is：
Wherein, X=(x1,x2,…,xn) it is optimal control vector, S is feasible zone or search space, RnIndicate n dimension
The space Euclidean, f (X) are objective function, gi(X) >=0 inequality constraints is indicated, K is the number of inequality constraints；
The objective function of above-mentioned optimization problem is rewritten as：
In formula, FmaxIt is the positive number to differ greatly with f (X) magnitude, is used for screening and is unable to satisfy the constrained trial solution of institute,
Consequently facilitating system carries out rejecting or other processing to it；
Memory elements are designed to data burst structure, including trial solution, memory residue, system time, memory state and forgetting shape
State, i.e.,
Mi=(Xi,mi,ti,si,fi), i=1,2 ..., N
In formula, point XiIndicate trial solution, N is memory elements number；miTo remember residue, memory journey of the characterization memory elements to stimulation
Degree, mi≥0；tiFor system time, that is, search for the time carried out；siFor memory state, there are I, tri- states of S, L, respectively wink
When, in short-term with long-term memory state；fiTo forget state, work as fiShow memory elements M when=1iTrial solution X is forgotteni, work as fi=0
When then indicate clearly to remember.
3. the aerial vehicle trajectory optimization method according to claim 2 based on artificial memory, it is characterised in that：The step
Two are specially：
Assuming that during the optimal solution search of the optimization problem, at current time (t), point XiProperty will receive other points
The influence of state, enables the set that other points are constituted when previous moment (t-1), i.e. trial solution collection is combined into
In order to guarantee global convergence and convergence rate, the trial solution search strategy is as follows：
1) Average Strategy
A certain number of memory elements are possessed in long-term memory system at this time, i.e., good trial solution；From wherein selecting at random
L trial solution out enables the trial solution at current time (t) for the average value of L trial solution in previous moment (t-1), i.e. point XiIt is right
The trial solution value answered is：
In formula,i1,i2,…,iLTo be taken from trial solution set X
The serial number of L trial solution out, n are point XiIn variable number；
2) orthogonal strategy
Two memory elements, serial number i are randomly selected from trial solution set X at this time1With i2, corresponding trial solution progress orthogonal weight
It operates, then point XiCorresponding trial solution value is：
In formula, a is the random number between 0~1, i.e. a=rand (0,1)；
3) projection strategy
Two memory elements, serial number i are randomly selected from trial solution set X at this time1With i2, corresponding trial solution progress projection operation,
Then point XiCorresponding trial solution value is：
4) expanding policy
A memory elements, serial number k are randomly selected from trial solution set X at this time, corresponding trial solution is extended operation, then point
XiCorresponding trial solution value is：
In formula, coefficient r1=rand (- 1,1)；
5) Chevy strategy
Choose the optimal trial solution of previous moment and the optimal trial solution at current time respectively from trial solution set X at this timeCorresponding trial solution carries out Chevy operation, then point XiCorresponding trial solution value is：
In formula, coefficient r2=rand (0.4,0.6), r3=1.3rand (0,1), r4=1.8rand (0,1)；
Above-mentioned 5 strategies can be all executed randomly in each search, each strategy after entire Optimizing Search process
The number being performed is of substantially equal.
4. the aerial vehicle trajectory optimization method according to claim 3 based on artificial memory, it is characterised in that：The step
Three are specially：
1) effective stimulus
For point XiIf compared with previous moment (t-1), target function value is reduced, i.e., in the search at current time (t)
Then illustrate the trial solution quality at current time better than previous moment, which is effective stimulus；
2) ineffective stimulation
Correspondingly, for point XiIf compared with previous moment (t-1), target function value is not in the search at current time (t)
Reduce, i.e.,
Then illustrate that the trial solution quality at current time has no improvement, which is ineffective stimulation.
5. the aerial vehicle trajectory optimization method according to claim 4 based on artificial memory, it is characterised in that：The step
Four are specially：
Forgetting modelling is carried out using Ebbinghaus forgetting curve model, then forgetting function is：
In formula, t is the time, since 0；Δ t is taken as regular time increment；α > 0 is to forget velocity coeffficient；
Known effective stimulus can be improved memory, and memory residue is enabled to increase, then enabling the increment of memory residue is Δ mi(t)；And
The absolute value of target function value incrementBigger, then the stimulation before showing the stimulation ratio is more effective, is intuitively presented as
Stimulation leads to the increment Delta m for remembering residuei(t) bigger；Then, using the memory residue increment Delta m of proportional relationshipi
(t) it is：
In formula, positive number h is effective stimulus proportionality coefficient, is taken as constant, for adjusting memory residue variation caused by stimulation greatly
It is small；
Remember at this time and has just been divided into two parts：The new memory increment that existing old memory and stimulation generate, then remembers residue
It is expressed as
mi(t+ Δ t)=mi(t)+Δmi(t)
In addition, memory should also include the memory part to old memory, according to Ebbinghaus forgetting curve it is found that the something lost of memory
Forgetting rule is speed difference that is fixed, only forgeing；
In order to distinguish old memory and to the memory of old memory, the memory residue at current time is expressed as form：
mi(t+ Δ t)=mi(t)e-αΔt+λmi(t)e-bΔt
In formula, function lambda mi(t)e-bΔtFor the memory value of memory, b is constant and b > α, that is, the forgetting speed recalled is faster；λ is one
A positive coefficient enables λ < 1, that is, recalling can not remain more than the memory of original memory for adjusting the forgetting speed recalled
Value；
Then above formula can be rewritten as
mi(t+ Δ t)=mi(t)(e-αΔt+λe-bΔt)
The links for comprehensively considering memory models, finally obtain the more new model of total recall, and expression formula is
mi(t+ Δ t)=mi(t)(e-αΔt+λe-bΔt)+Δmi(t)
Enable β=e-αΔt+λe-bΔt, when time increment Delta t is fixed value, then β is also fixed constant, then remembers more new model
It is reduced to following form：
In formula, factor betaI,βS,βLIt is respectively instantaneous, in short-term with the forgetting speed of long-term memory, by remembering in three memory systems
Forgetting speed it is found that 0 < βL< βS< βI< 1.
6. the aerial vehicle trajectory optimization method according to claim 5 based on artificial memory, it is characterised in that：The step
Five are specially：
Wherein the setting of parameters is as follows：
1) incremental time Δ t=0.1, system sweeps number or maximum number of iterations is G=10000；
2) element number of optimal control vector X, i.e. n=100；
3) memory elements number N=100, the memory elements number L=3 selected in long-term memory system；
4) effective stimulus proportionality coefficient h=1.2；
5) instantaneous, in short-term with the forgetting velocity coeffficient β of long-term memoryI=0.55, βS=0.35, βL=0.15.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810768925.9A CN108919828A (en) | 2018-07-13 | 2018-07-13 | A kind of aerial vehicle trajectory optimization method based on artificial memory |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810768925.9A CN108919828A (en) | 2018-07-13 | 2018-07-13 | A kind of aerial vehicle trajectory optimization method based on artificial memory |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108919828A true CN108919828A (en) | 2018-11-30 |
Family
ID=64412585
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810768925.9A Pending CN108919828A (en) | 2018-07-13 | 2018-07-13 | A kind of aerial vehicle trajectory optimization method based on artificial memory |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108919828A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114397894A (en) * | 2021-12-29 | 2022-04-26 | 杭州电子科技大学 | Mobile robot target searching method simulating human memory |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103995540A (en) * | 2014-05-22 | 2014-08-20 | 哈尔滨工业大学 | Method for rapidly generating finite time track of hypersonic aircraft |
| CN104615144A (en) * | 2015-01-30 | 2015-05-13 | 天津大学 | Goal programming based hypersonic flight vehicle re-entry trajectory online optimization method |
-
2018
- 2018-07-13 CN CN201810768925.9A patent/CN108919828A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103995540A (en) * | 2014-05-22 | 2014-08-20 | 哈尔滨工业大学 | Method for rapidly generating finite time track of hypersonic aircraft |
| CN104615144A (en) * | 2015-01-30 | 2015-05-13 | 天津大学 | Goal programming based hypersonic flight vehicle re-entry trajectory online optimization method |
Non-Patent Citations (2)
| Title |
|---|
| GUANG-QIU HUANG: "Artificial memory optimization", 《APPLIED SOFT COMPUTING》 * |
| 周子源: "高超声速飞行器助推段轨迹优化与制导技术研究", 《中国优秀硕士学位论文全文数据库工程科技Ⅱ辑》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114397894A (en) * | 2021-12-29 | 2022-04-26 | 杭州电子科技大学 | Mobile robot target searching method simulating human memory |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109615011A (en) | A kind of middle and small river short time flood forecast method based on LSTM | |
| CN111009134A (en) | Short-term vehicle speed working condition real-time prediction method based on interaction between front vehicle and self vehicle | |
| Boda et al. | Stochastic target hitting time and the problem of early retirement | |
| CN107578061A (en) | Based on the imbalanced data classification issue method for minimizing loss study | |
| CN104881582B (en) | The Forecasting Methodology and device of air quality | |
| CN109711534A (en) | Dimensionality reduction model training method, device and electronic equipment | |
| CN111967696B (en) | Neural network-based electric vehicle charging demand prediction method, system and device | |
| CN111612125A (en) | Novel HTM time pool method and system for online learning | |
| CN107016212B (en) | Intention analysis method based on dynamic Bayesian network | |
| CN114167898B (en) | Global path planning method and system for collecting data of unmanned aerial vehicle | |
| CN107403222A (en) | A kind of motion tracking method based on auxiliary more new model and validity check | |
| CN109214499A (en) | A kind of difference searching algorithm improving optimizing strategy | |
| Zhang et al. | Short‐term load forecasting based on wavelet neural network with adaptive mutation bat optimization algorithm | |
| CN110427690A (en) | A kind of method and device generating ATO rate curve based on global particle swarm algorithm | |
| CN108919828A (en) | A kind of aerial vehicle trajectory optimization method based on artificial memory | |
| CN103870548B (en) | The access control method of spatial database | |
| CN108181809B (en) | System error-based parameter self-tuning method for MISO (multiple input single output) compact-format model-free controller | |
| CN106650028A (en) | Optimization method and system based on agile satellite design parameters | |
| Ramsey et al. | Simulation-assisted learning by competition: Effects of noise differences between training model and target environment | |
| CN109115446A (en) | Based on transonic wind tunnel wind speed accuracy control method and system | |
| CN108288074A (en) | A kind of selection method and system of data characteristics | |
| CN106919955A (en) | A kind of two points of K mean algorithms based on density criteria for classifying | |
| CN105677936B (en) | The adaptive recurrence multistep forecasting method of electromechanical combined transmission system demand torque | |
| CN103927776B (en) | animation curve optimization method | |
| CN108763604A (en) | A kind of radial base neural net point collocation solving the response of the statics of composite structure containing interval parameter |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181130 |