CN113033754B - Evaluation method for high-frequency ground wave radar target tracking algorithm based on collaborative scene evolution - Google Patents
Evaluation method for high-frequency ground wave radar target tracking algorithm based on collaborative scene evolution Download PDFInfo
- Publication number
- CN113033754B CN113033754B CN202011422657.9A CN202011422657A CN113033754B CN 113033754 B CN113033754 B CN 113033754B CN 202011422657 A CN202011422657 A CN 202011422657A CN 113033754 B CN113033754 B CN 113033754B
- Authority
- CN
- China
- Prior art keywords
- population
- evolution
- parameter
- parties
- individual
- 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.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/004—Artificial life, i.e. computing arrangements simulating life
- G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/12—Computing arrangements based on biological models using genetic models
- G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming
Abstract
The invention relates to a high-frequency ground wave radar target tracking algorithm evaluation method based on collaborative scene evolution, and belongs to the technical field of high-frequency ground wave radars. The invention collaborates two parties by initialization: generating an initial population according to respective coding modes of the test scene population and the parameter configuration population; calculating the proper value of the two cooperative parties: the parameter proper value is the number of scenes successfully tracked under the parameter configuration, and the scene proper value is the number of times of unsuccessfully tracked during the configuration of each parameter; competitive co-evolution based on equilibrium mechanism variants: determining an evolved population according to a balance mechanism of adding a threshold protection inferior population strategy and a strong population degradation strategy, wherein genetic programming is applied to the evolution of the situation population, and a differential evolution algorithm is applied to the evolution of the parameter population. The method can realize the failure situation with difficult tracking and the coevolution of the optimized configuration parameters of the target tracking algorithm, thereby obtaining the optimized configuration set of the target tracking algorithm while discovering the essential defects of the target tracking algorithm.
Description
Technical Field
The invention relates to a high-frequency ground wave radar target tracking algorithm evaluation method based on collaborative scene evolution, and belongs to the technical field of high-frequency ground wave radars.
Background
At present, competitive coevolution has been successfully applied to many fields such as function optimization, image recognition, real-time strategy games and the like, and compared with the traditional evolutionary computation, the competitive coevolution algorithm has the advantage of effectively relieving the problem of evolutionary stagnation in the complex optimization, and is a hotspot research content in the field of the current evolutionary computation. The competitive coevolution algorithm is used for evaluation of the high-frequency ground wave radar target tracking algorithm, and the evaluation is promising: the traditional single group evolution can only obtain complex test scenes under the tracking algorithm of fixed parameters, and the cooperative evolution of failure scenes with difficult tracking and optimized configuration parameters of the target tracking algorithm can be realized by applying competitive cooperative evolution, so that the optimized configuration set of the target tracking algorithm is obtained while the essential defects of the target tracking algorithm are found. One party of competition is a test scenario, the other party of competition is parameter configuration of an algorithm, the performance of the challenge algorithm is used for testing scenario population evolution, the population evolution optimization algorithm is configured according to the parameters so as to respond to the challenge, finally, the essential defect of the algorithm is obtained by one party (scenario population) of competition, the space where the algorithm can be improved is found by the other party (parameter configuration population) in the evolution process, the optimization configuration set of the algorithm under various different scenarios is determined, the organic combination of discovery and repair of failure scenarios can be realized, the two parties are mutually promoted, and the effect is dynamically improved.
However, it should be noted that there are still some difficulties in applying the existing competitive coevolution algorithm to the evaluation of the high-frequency ground wave radar target tracking algorithm. The key of the effective operation of the competitive coevolution algorithm lies in maintaining the competition of two parties, so that the two parties have equal force, thereby avoiding the unbalance of competition and the stagnation of evolution, and in the evaluation application of the high-frequency ground wave radar target tracking algorithm, two problems that the evolution is possibly caused to be trapped in stagnation exist: firstly, the search space of the test scene population is far larger than the parameter configuration, the difference of the evolution search efficiency of the two parties in the co-evolution is large, and the evolution speeds of the two parties in the co-evolution are inconsistent; secondly, the essential defect of the algorithm is that the situation population is a strong individual, so that the competitor-algorithm parameter configuration population completely loses the competitiveness, and the competition between the two parties of the co-evolution is permanently unbalanced. Therefore, the existing competitive coevolution algorithm is improved to solve the problems in the evaluation application of the high-frequency ground wave radar target tracking algorithm.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a high-frequency ground wave radar target tracking algorithm evaluation method based on collaborative scene evolution, which is used for generating various scenes to test a target tracking algorithm, so that on one hand, scenes causing algorithm failure are searched, and intrinsic defects of the algorithm are distinguished from the scenes, on the other hand, an improved space of the algorithm is distinguished, an optimized configuration set of the algorithm is provided, and a high-efficiency test evaluation method and a decision support tool are provided for establishing a perfect high-frequency ground wave radar target monitoring and tracking system.
The invention discloses a high-frequency ground wave radar target tracking algorithm evaluation method based on collaborative scene evolution, which comprises the following steps of:
s1: initialization of the two parties in cooperation: namely, the generation of the test scene and the parameter configuration comprises the test scene population and the parameter configuration population initializing the populations according to respective coding modes, wherein:
s11: the method comprises the following steps that a test scenario is constructed based on a parameterized context-free grammar model, defined grammar rules of limited quantity are used for describing the motion situation of a naval vessel, and a test scenario population generates scenarios of different motion situations of a certain quantity of naval vessel targets in a detected sea area;
s12: the parameter configuration method comprises the steps that random numbers are generated and combined into real number arrays to serve as individuals of a parameter population according to respective value ranges of selected key parameters through parameter configuration, and the parameter configuration population generates a certain number of parameter arrays which are selected according to a target tracking algorithm to be detected and influence the performance of the algorithm;
s2: calculating the appropriate value of the two parties in cooperation: setting a high-frequency ground wave radar target tracking algorithm according to parameter individuals in a parameter configuration population, taking scene individuals in a test scene population as input of the target tracking algorithm, running the tracking algorithm to obtain a tracking result, setting an evaluation index and a calculation mode thereof, defining that the target tracking algorithm under the parameter setting completes tracking of the test scene when an evaluation value is in a certain range, defining the fitness value of the scene individuals as the number of times of incapability of tracking, and defining the fitness value of the parameter individuals as the number of times of successful completion of tracking;
s3: competitive co-evolution based on equilibrium mechanism variants: evaluating the population strength of the two parties according to the fitness conditions of the two parties in cooperation so as to achieve the purposes of keeping competition of the two parties in competition and avoiding stagnation of evolution, only evolving one population of the two parties in cooperation for each generation, selecting the population to be evolved for the generation according to the cooperation strategy set by a balance mechanism and variants thereof, and determining the evolution content of the two parties in cooperation at present;
s4: and (3) the two parties collaborate to evolve according to requirements: and (4) evolving the selected cooperative party according to the decision obtained in the S3, wherein the situation population applies an evolutionary method of genetic programming, and the parameter population applies an evolutionary method of differential evolution.
Preferably, in step S3, the competitive coevolution algorithm based on the equilibrium mechanism variant can keep the competition of the two competing parties and avoid the stagnation of evolution when the evolutionary search efficiencies of the two competing parties are inconsistent or one of the competing parties appears in a strong individual, and specifically includes the following steps:
s31: aiming at the obtained fitness values of the individuals of the two competing parties, reflecting the population strength of the two competing parties at present by defining a plurality of population strength measuring modes, and obtaining respective reproduction rates of the two competing parties according to the population strengths of the two competing parties, namely the probability that various populations are selected for evolution;
s32: aiming at the problems of inconsistent evolution search efficiency and large evolution speed difference of two competing parties, a balance mechanism is applied to solve the problems, namely, only one population is evolved in each generation in the specified evolution process, the probability that various populations are selected to be evolved, namely the reproduction rate, is determined by the population momentum strength and is inversely proportional to the population momentum strength, and the lower the population reproduction rate with stronger momentum, the smaller the probability of selected evolution;
wherein, the reproduction rate of various groups is inversely proportional to the strength of the potential force of the groups, and the strength of the potential force of the scene groups is assumed to be Str S The force intensity of the parameter population is 1-Str S Then the reproduction rate of the episodic population is 1-Str S The breeding rate of the parameter population is Str S ；
S33: according to a threshold value setting strategy which is set by an algorithm and aims at protecting the inferior population, judging whether the reproduction rate of the situation population reaches a threshold value, and if the reproduction rate does not reach the threshold value, directly judging to select the situation population for evolution;
s34: a degradation strong population strategy which is set according to an algorithm and aims to solve the problem that the situation that a strong individual appears in one population possibly appearing in the evolution process and the other population completely loses competitiveness is as follows: judging whether the potential force difference between the two populations is overlarge, and when the potential force difference between the two populations is overlarge, carrying out degeneration operation on the strong party, namely changing the fitness condition of the individuals, so that the individuals with backward ranking in the populations are selected to be evolved in the subsequent evolution selection process, thereby achieving the purpose of reducing the strength of the strong populations;
the degradation strong population strategy is characterized in that the fitness value of an individual in a strong population is recalculated, so that the individual which is originally in a low fitness value behind the ranking in the population obtains a higher fitness value, and is selected to participate in generation of filial generations with a higher probability, thereby reducing the overall strength of the strong population, balancing the two populations, keeping competition of two parties in competition and avoiding stagnation of evolution;
the fitness recalculation is as follows:
f′＝2ω 2 x-ωx 2
wherein f' is the recalculated individual fitness value, x is the score value of each individual in the interval of [0,1] obtained by normalizing the fitness values f of all the individuals of the dominant population, and ω is the degradation coefficient in the interval (0, 0.5) representing the degradation degree of the population, which is determined by the difference between the populations, wherein the larger the difference is, the smaller the degradation coefficient is, and the higher the degradation degree is.
Preferably, in step S2, a test scenario is tested under a target tracking algorithm corresponding to a parameter, and the degree of track fracture is selected as an evaluation index of the tracking condition: when the track fracture degree exceeds a certain range, it is indicated that the target tracking algorithm under the parameter setting cannot track the test scenario successfully.
Preferably, in step S3, the current evolution situation is determined according to the algorithm requirement and the current competitive situation: determining a population to be evolved according to the population reproduction rate and a weak population protection strategy for setting a threshold; according to the strategy of the degeneration strong population, when the inter-species difference is too large, the current generation is not normally evolved, but the strong population is selected to perform the evolution operation towards the degeneration direction, and the proper value of the individual in the current strong population needs to be recalculated.
Preferably, in step S31, the population potential strength is calculated by two methods, namely, by race calculation and by fitness average calculation:
the first method is as follows: the competition calculation mode is that one individual is selected from the competition parties, the appropriate value of the individual is compared, the individual with the large appropriate value wins, a certain number of competitions are carried out, and the ratio of the winning times of the competition parties to the total number of competitions is obtained, namely the strength of each momentum;
the second method comprises the following steps: the mode of calculating the mean value of the fitness values of the two groups of the competition parties is respectively calculated, and the strength of each group potential force is the proportion of the mean value to the sum of the mean values.
Preferably, in the step S33, the disadvantage population is a population with a slow evolution speed in the evolution process, and is a situation population; in step S34, the degeneration-strong population is a population that is too powerful and completely defeats the adversary state.
Preferably, in step S4, when selecting the scenario population to evolve or performing a degeneration operation on the scenario population as a strong population, the scenario population coded by the grammar deduction tree adopts genetic programming evolution based on grammar guidance; when selecting the parameter populations, a differential evolution method is employed for the parameter populations encoded by the real number array.
Preferably, in step S4, when selecting the parameter group, a differential evolution method is applied to the parameter group encoded by the real number array: the difference mutation operator is shown as follows:
wherein the content of the first and second substances,
is a new individual obtained by mutation operation of the ith individual of the G generation,
represents the ith individual of the G-th generation,
for the first p% of the best individualsR1 and r2 are any two individuals in the G generation population, and F is a scaling factor;
the differential crossover operator is shown as follows:
wherein the content of the first and second substances,
is a new individual obtained by the ith individual of the G generation through the crossover operation, wherein Cr is the crossover rate, j is rand An integer defining a random position selected for each individual;
the difference algorithm applies a greedy selection strategy and hopes to retain the individuals with high fitness to the next generation, namely when the fitness of the new individual obtained after the variation crossing operation is larger than that of the old individual, the new individual is retained to the next generation, otherwise, the old individual is retained to the next generation, as shown in the following formula:
the invention has the beneficial effects that:
(1) by applying competitive coevolution, the inefficacy situation with difficult tracking and the coevolution of the optimized configuration parameters of the target tracking algorithm are realized, so that the optimized configuration set of the target tracking algorithm is obtained while the essential defect of the target tracking algorithm is found;
(2) by the designed competitive coevolution algorithm based on the equilibrium mechanism variant, the problems that the evolution search efficiency difference of two parties of coevolution is large and the evolution speeds of the two parties of coevolution are inconsistent are effectively solved;
(3) by the designed competitive coevolution algorithm based on the balance mechanism variant, the essential defects of algorithms such as 'strong individuals' of the situation population which may occur are effectively overcome, the algorithm parameter configuration population such as a competitor of the algorithm parameter configuration population completely loses competitiveness, and the competition between the two parties of coevolution is permanently unbalanced.
Drawings
FIG. 1 is a schematic of the architecture of the present invention;
FIG. 2 is an overall flow diagram of the present invention;
3(a) -3 (b) are schematic diagrams of test scenarios and parameter configuration designs;
4(a) -4 (b) are schematic diagrams of crossover operator, mutation operator design for test scenarios;
FIG. 5 is a schematic diagram of a differential mutation operator design for a parameter configuration;
fig. 6(a) -6 (c) are diagrams illustrating the results of the implementation operation during initialization, during evolution, and after evolution is completed.
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 obtained by a person skilled in the art without making any creative effort based on the embodiments in the present invention, belong to the protection scope of the present invention.
Example 1:
as shown in fig. 1 to fig. 6(c), the method for evaluating a high-frequency ground wave radar target tracking algorithm based on collaborative scene evolution of the present invention includes the following steps:
s1: initialization of both parties in cooperation, namely generation of test scenarios and parameter configuration: the method comprises the steps that a test scene population and a parameter configuration population initialize populations according to respective coding modes, the test scene is constructed on the basis of a parameterized context-free grammar model, defined finite grammar rules are used for describing the motion conditions of a marine ship, the test scene population generates a certain number of scenes of different motion situations of ship targets in a detected sea area, the parameter configuration aims at selected key parameters, random numbers are generated according to respective value ranges of the selected key parameters and combined into a real number group to be used as an individual of the parameter population, and the parameter configuration population generates a certain number of parameter groups which influence algorithm performance and are selected by a target tracking algorithm to be detected;
s2: calculating the appropriate value of the two parties in cooperation: setting a high-frequency ground wave radar target tracking algorithm according to parameter individuals in a parameter configuration population, taking scene individuals in a test scene population as input of the target tracking algorithm, running the tracking algorithm to obtain a tracking result, setting an evaluation index and a calculation mode thereof, defining that the target tracking algorithm under the parameter setting can finish tracking the test scene when an evaluation value is in a certain range, defining the fitness value of the scene individual as the number of times that the scene individual cannot be tracked, and defining the fitness value of the parameter individual as the number of times that the tracking is successfully finished;
s3: competitive co-evolution based on equilibrium mechanism variants: evaluating the population strength of two parties according to the fitness conditions of the two parties in cooperation, in order to realize the purposes of keeping competition between the two parties and avoiding stagnation of evolution, only one population of the two parties in cooperation is evolved in each generation, the population to be evolved is selected for the generation according to the cooperation strategy set by a balance mechanism and variants thereof, and the evolution content to be performed by the two parties in cooperation at present is determined;
s4: and (3) the two parties collaborate to evolve according to requirements: and (4) evolving the selected cooperative party according to the decision obtained in the S3, wherein the situation population applies an evolutionary method of genetic programming, and the parameter population applies an evolutionary method of differential evolution.
Preferably, in S2, a test scenario is tested under a target tracking algorithm corresponding to an individual parameter, and the degree of track fracture is selected as an evaluation index of the tracking condition: when the track fracture degree exceeds a certain range, it is indicated that the target tracking algorithm under the parameter setting cannot track the test scenario successfully.
Preferably, in S3, the competitive collaborative algorithm based on the equilibrium mechanism variant can keep the competition of the two competing parties and avoid the stagnation of evolution when the evolution search efficiency of the two competing parties is inconsistent or one of the competing parties appears in a strong individual, specifically including the following steps:
s31: aiming at the obtained fitness values of the individuals of the two competing parties, reflecting the strength of the population potentials of the two competing parties at present by defining a plurality of population potential measuring modes, and obtaining respective reproduction rates of the two competing parties, namely the probability of the evolution of various groups selected according to the population potentials of the two competing parties;
s32: aiming at the problems of inconsistent evolution search efficiency and large evolution speed difference of two competing parties in the invention, a balance mechanism is applied to solve the problems, namely, only one population is evolved in each generation in the specified evolution process, the probability that various populations are selected to be evolved, namely the reproduction rate, is determined by the population strength and is inversely proportional to the population strength, and the lower the population reproduction rate with stronger potential is, the smaller the probability of selected evolution is;
s33: according to a strategy of setting a threshold value set by an algorithm and aiming at protecting a disadvantage population (namely a population with a low evolution speed in the evolution process, namely a situation population in the invention), judging whether the reproduction rate of the situation population reaches the threshold value or not, and if not, directly judging to select the situation population for evolution;
s34: a degradation strong population strategy (i.e. a population in an over-strong and complete defeat adversary state) which is set according to an algorithm and aims to solve the situation that a strong individual appears in a population of one party possibly appearing in the evolution process and the population of the other party completely loses competitiveness: and judging whether the difference of the forces between the two populations is overlarge, and when the difference of the forces between the two populations is overlarge, performing degeneration operation on the strong party, namely changing the fitness condition of the individuals, so that the individuals with backward ranking in the populations are selected to evolve in the subsequent evolution selection process, thereby achieving the purpose of reducing the strength of the strong populations.
Preferably, in S31, the population potential strength is calculated by two methods, namely, by race calculation and by fitness average calculation: the competition calculation mode is that one individual is selected from the competition parties, the appropriate value of the individual is compared, the individual with the large appropriate value wins, a certain number of competitions are carried out, and the ratio of the winning times of the competition parties to the total number of competitions is obtained, namely the strength of each momentum; the mode of calculating the mean value of the fitness values of the two groups of the competition parties is respectively calculated, and the strength of each group potential force is the proportion of the mean value to the sum of the mean values.
Preferably, in S32, various groups are propagatedThe rate is inversely proportional to the population potential strength, assuming that the potential strength of the scene population is Str S The force intensity of the parameter population is 1-Str S Then the reproduction rate of the episodic population is 1-Str S The breeding rate of the parameter population is Str S 。
Preferably, in S34, the degradation dominant population strategy is to recalculate the fitness value of the individual in the dominant population, so that the individual originally in the population with a low fitness value behind the ranking obtains a higher fitness value, and a greater probability is selected to participate in generating offspring, thereby reducing the overall strength of the dominant population, balancing the two populations, and thus keeping competition between the two competing parties and avoiding stagnation of evolution. The formula for the fitness recalculation is shown as follows:
f′＝2ω 2 x-ωx 2
wherein f' is the recalculated individual fitness value, x is the score value of each individual in the interval of [0,1] obtained by normalizing the fitness values f of all the individuals of the dominant population, and ω is the degradation coefficient in the interval (0, 0.5) representing the degradation degree of the population, which is determined by the difference between the populations, wherein the larger the difference is, the smaller the degradation coefficient is, and the higher the degradation degree is.
Preferably, in S3, the current evolution situation is determined according to the algorithm requirement and the current competitive situation: determining a population to be evolved according to the population reproduction rate and a weak population protection strategy for setting a threshold; according to the strategy of the degeneration strong population, when the inter-species difference is too large, the current generation is not normally evolved, but the strong population is selected to perform the evolution operation towards the degeneration direction, and the proper value of the individual in the current strong population needs to be recalculated.
Preferably, in S4, when selecting the episodic population to evolve or performing the degeneration operation on the episodic population as the episodic population, the episodic population encoded by the grammar deduction tree is subjected to genetic programming evolution based on grammar guidance: the selection operator uses the championship selection, the parameter and the number contained in the two subtrees and subtrees of the crossover operator and the mutation operator to select two nodes of the same tree and exchange the two nodes and the subtrees.
Preferably, in S4, when selecting the parameter group, a differential evolution method is applied to the parameter group encoded by the real number array: the difference mutation operator is shown as follows:
wherein
Is a new individual obtained by mutation operation of the ith individual of the G generation,
represents the ith individual of the G-th generation,
r1 and r2 are any two individuals in the G generation population, and F is a scaling factor; the differential crossover operator is shown below:
wherein
Is a new individual obtained by the ith individual of the G generation through the crossing operation, Cr is the crossing rate, j is rand An integer defining a certain random position selected for each individual;
the difference algorithm applies a greedy selection strategy and hopes to retain the individuals with high fitness to the next generation, namely when the fitness of the new individual obtained after the variation crossing operation is larger than that of the old individual, the new individual is retained to the next generation, otherwise, the old individual is retained to the next generation, as shown in the following formula:
the principle of the invention is as follows: the method comprises the steps that a competitive coevolution algorithm based on a balance mechanism variant is used, different coding modes and evolutionary algorithms are adopted for parameter configuration of a high-frequency ground wave radar marine target tracking test scene and a target tracking algorithm, a test scene population adopts grammar-based genetic programming evolution, and a parameter configuration population adopts a differential evolution algorithm for evolution; according to a competitive coevolution algorithm based on a balance mechanism variant, a balance method for setting a threshold value and a strong population degradation strategy are applied, so that the problems that search space difference is large, evolution speed is inconsistent, and a strong individual appears in one of test scenes which may appear in the evolution process, so that a parameter configuration population completely loses competitiveness, and competition between two coevolution parties is permanently unbalanced are solved; by applying competitive coevolution, the coevolution of failure scenes with difficult tracking and optimized configuration parameters of the target tracking algorithm can be realized, so that the optimized configuration set of the target tracking algorithm is obtained while the essential defects of the target tracking algorithm are found.
By applying competitive coevolution, the invention can realize the inefficacy situation with difficult tracking and the coevolution of the optimized configuration parameters of the target tracking algorithm, thereby obtaining the optimized configuration set of the target tracking algorithm while discovering the essential defects of the target tracking algorithm; the competitive coevolution algorithm based on the equilibrium mechanism variant effectively solves the problems that the evolution search efficiency difference of both parties of coevolution is large and the evolution speeds of both parties of coevolution are inconsistent, which are possibly generated in the invention; the competitive coevolution algorithm based on the balance mechanism variant effectively solves the essential defects of 'strong individual' of the situation population-the algorithm which possibly appears in the invention, and can lead the competitor-the algorithm parameter configuration population to completely lose the competitiveness and the competition between the coevolution parties to be permanently unbalanced.
Example 2:
the present invention will be described in further detail with reference to specific examples.
As shown in fig. 1, the architecture diagram of the embodiment of the present invention specifically includes the following steps:
s1: initializing (100) the test scenes and parameter configurations of the two cooperative parties, respectively generating the test scenes (101) of the high-frequency ground wave radar marine target tracking algorithm and the parameter configurations (102) of the target tracking algorithm, simulating the motion scenes of five to fifteen ships in the sea area, and setting reasonable values for key parameters of the target tracking algorithm.
S2: calculating the fitness values of the two cooperative parties (103), operating to obtain a tracking result of the target by taking the generated parameters as the parameter configuration of the target tracking algorithm and the generated test scenes as the input of the target tracking algorithm, on the basis of the result, evaluating the tracking performance by taking the fracture degree of the tracking track as an index (104), evaluating whether the test scenes are successfully tracked according to the calculated performance evaluation value, respectively calculating the fitness value (105) of the test scenes and the fitness value (106) of the parameter configuration, adding 1 to the fitness value of the parameter configuration of the algorithm if the tracking is successful, and adding 1 to the fitness value of the test scenes if the tracking is not successful, namely, the fitness value of the test scenes is the number of times that the target tracking algorithm under the parameter configuration successfully tracks and the fitness value of the parameter configuration is the number of the test scenes that the algorithm successfully tracks under the configuration.
S3: a competitive coevolution algorithm (107) based on equilibrium mechanism variants is applied, the interspecies potential strength is evaluated (108) according to the calculated fitness conditions of both cooperative parties, and the population to be evolved and the direction of evolution of the current generation are determined according to the algorithm requirements for the potential strength of the two populations (109).
In a specific embodiment: the strength of the momentum is evaluated by selecting a competition mode, an individual is respectively selected from two competing populations, the fitness values of the individual and the population are compared, the one with the larger fitness value wins, the competition is repeatedly carried out for a plurality of times, and the respective winning rate of the two populations is calculated to be used as the strength of the momentum of the populations.
In a specific embodiment: a competitive coevolution algorithm based on a balance mechanism variant is provided to maintain the balance of the forces among the populations, and a weak population protection strategy and a strong population degradation strategy for setting a threshold value are added to the algorithm on the basis of a balance method. The algorithm stipulates that only one population is selected for evolution in each generation, and the probability of various population evolutions, namely the reproduction rate, is formed by the population strengthInversely proportional, assuming the strength of the force of the situational population is Str S The force intensity of the parameter population is 1-Str S Then the reproduction rate of the episodic population is 1-Str S The breeding rate of the parameter population is Str S 。
According to the weak population protection strategy requirement for setting the threshold, when the reproduction rate of the situation population is lower than the threshold, considering that the search space of the situation population evolution far exceeds the parameter population evolution space, the evolution speed is slow, the reproduction rate of the situation population is directly set to be 1, the reproduction rate of the parameter population is 0, namely, the situation population is directly selected for evolution, and the situation that the parameter population is continuously evolved and strengthened under the condition of strong strength of the potential is avoided.
According to the strategy requirement of the degradation strong population, when the strength difference of the two population potentials is large, the suitable value of the individual in the strong population is recalculated by the strong population, so that the individual which is originally in the low suitable value behind the ranking in the population obtains a higher suitable value, and more probability is selected to participate in generation of filial generation, thereby reducing the overall strength of the strong population, balancing the potentials of the two populations, keeping competition of the two populations, and avoiding stagnation of evolution. The formula for the fitness recalculation is shown as follows:
f′＝2ω 2 x-ωx 2
wherein f' is the recalculated individual fitness value, x is the score value of each individual in the interval of [0,1] obtained by normalizing the fitness values f of all the individuals of the dominant population, and omega is the degradation coefficient in the interval (0, 0.5) and represents the degradation degree of the population, and is determined by the difference between the populations, and the larger the difference is, the smaller the degradation coefficient is, the higher the degradation degree is.
S4: and (2) the two parties are collaboratively evolved according to the requirement (110), an evolved population and an evolution direction are selected according to the evolution content obtained by the algorithm requirement in the last step, and the situation population is evolved or degenerated (111) or the parameter population is evolved or degenerated (112).
In a specific embodiment: the episodic population is evolved (111) by grammar-based genetic programming (GGGP), and the parametric population is evolved (112) by a differential evolution algorithm (DE).
S5: and (5) circulating the steps 2, 3 and 4 until the termination condition is reached.
FIG. 2 shows a competitive coevolution flow chart of a high-frequency ground wave radar target tracking algorithm test scenario and parameter configuration based on a balance mechanism variant. And (2) carrying out fitness calculation (202) on the situation population (200) and the parameter population (201) generated by initialization, evaluating the strength of the two populations according to fitness conditions, determining the population to be evolved (203) according to algorithm requirements, judging whether the difference between the two competing populations is too large (204), and degrading the strong one when the difference is larger (205). And repeating the evolution process, continuously updating to generate a new population by utilizing a competition mechanism of competitive co-evolution and a self-feedback mechanism implied by evolutionary computation, and obtaining an optimized configuration set of the target tracking algorithm while discovering the essential defects of the target tracking algorithm.
Fig. 3(a) -3 (b) show a test scenario and parameter configuration design diagram of the present invention, where fig. 3(a) is a test scenario design diagram, and the test scenario employs modeling based on a parameterized context-free grammar, and first generates a grammar "deduction tree" (301) of a scenario composed of grammar rule sequence numbers according to grammar rules, then replaces the deduction tree with the content of the rules, and represents the event content represented by the rule numbers to obtain a "scenario frame" (302), and finally reads the motion situation of each ship target in the scenario frame to obtain a "scenario string" (303), and assigns values to each parameter in the string to generate a test scenario. Fig. 3(b) is a schematic diagram of parameter configuration design, where the parameters are represented in a real number array (304), and each element of the parameter array is an algorithm key parameter (205).
Fig. 4(a) -4 (b) are schematic diagrams illustrating genetic operator design of the test scenario of the present invention, and fig. 4(a) is a schematic diagram illustrating crossover operator design, wherein the crossover operator exchanges parameters and numbers contained in two subtrees simultaneously. FIG. 4(b) is a schematic diagram of mutation operator design, using subtree crossover mutation operators to select two nodes in an individual, crossover nodes and their subtrees.
In a specific embodiment: the father 1(400) and the father 2(401) are two father individuals selected to be crossed and respectively correspond to a scene frame 1(402) and a scene frame 2(403), and the inner frame part (404) in the father 1 and the inner frame part (405) in the father 2 are subtrees to be crossed and respectively correspond to the inner frame part (406) in the scene frame 1 and the inner frame part (407) in the scene frame 2. Child 1(408) and child 2(409) are child individuals obtained through crossing and respectively correspond to scenario frame 1(410) and scenario frame 2(411), and the intra-frame part (412) in child 1 and the intra-frame part (413) in child 2 are subtrees after crossing and respectively correspond to the intra-frame part (414) in scenario frame 1 and the intra-frame part (415) in scenario frame 2. The invention applies a crossover operator which, while exchanging two subtrees, also exchanges the parameters and numbers contained in the subtrees, which is determined according to the properties based on the parameterized context-free grammar modeling.
In a specific embodiment: two nodes to be switched are selected in the deduction tree (416) of the individuals before mutation, and a node 1 and a sub-tree thereof (418) and a node 2 and a sub-tree thereof (419) are selected, corresponding to a left-side scene frame (417) and scene frame intra-frame portions 1(420) and 2 (421). The switching nodes and their subtrees, as well as the parameters and numbers contained in the subtrees, result in mutated individuals (422) and their corresponding context frames (423). After mutation, nodes 1(425) and 2(424) exchange locations, and corresponding scenario framework intra-frame part 1(427) and intra-frame part 2(426) also exchange locations and parameters.
Fig. 5 shows a schematic diagram of designing a differential mutation operator for parameter configuration according to the present invention, which performs a differential operation on an individual (500) to be subjected to differential mutation, a randomly selected individual 1(501), an individual 2(502), and an optimal individual (503) in a population to obtain a new individual (503) after the differential mutation.
Fig. 6(a) -6 (c) show graphs of implementation operation results of the present invention, which respectively show high-frequency ground wave radar marine target tracking simulation test scenarios during initialization (600), during evolution (601), and after evolution (602).
In a specific embodiment: the part circled by the circle is the part with track fracture, the number is the number of the marine ship target in the test scene, and the moving direction of the ship target is driven from one end marked by the number to the other end.
Through the continuous iterative evolution of the cooperation of the test scenario and the parameter configuration, the failure scenario with difficult tracking and the target tracking algorithm configuration parameters are continuously optimized, and finally, the optimization configuration set of the target tracking algorithm can be obtained while the essential defects of the target tracking algorithm are found.
The invention can be widely applied to high-frequency ground wave radar occasions.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.
Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.
Claims (2)
1. A high-frequency ground wave radar target tracking algorithm evaluation method based on collaborative scene evolution is characterized by comprising the following steps:
s1: initialization of the two parties in cooperation: namely, the generation of the test scene and the parameter configuration comprises the test scene population and the parameter configuration population initializing the populations according to respective coding modes, wherein:
s11: the method comprises the following steps that a test scenario is constructed based on a parameterized context-free grammar model, defined grammar rules of limited quantity are used for describing the motion situation of a naval vessel, and a test scenario population generates scenarios of different motion situations of a certain quantity of naval vessel targets in a detected sea area;
s12: the parameter configuration aims at the selected key parameters, random numbers are generated and combined into a real number array to serve as individuals of a parameter population according to respective value ranges of the key parameters, and the parameter configuration population generates a certain number of parameter arrays which are selected aiming at the target to be detected and influence the performance of the algorithm;
s2: calculating the appropriate value of the two parties in cooperation: setting a high-frequency ground wave radar target tracking algorithm according to parameter individuals in a parameter configuration population, taking scene individuals in a test scene population as input of the target tracking algorithm, running the tracking algorithm to obtain a tracking result, setting an evaluation index and a calculation mode thereof, defining that the target tracking algorithm under the parameter setting completes tracking of the test scene when an evaluation value is in a certain range, defining the fitness value of the scene individuals as the number of times of incapability of tracking, and defining the fitness value of the parameter individuals as the number of times of successful completion of tracking;
s3: competitive co-evolution based on equilibrium mechanism variants: evaluating the population strength of the two parties according to the fitness conditions of the two parties in cooperation so as to achieve the purposes of keeping competition of the two parties in competition and avoiding stagnation of evolution, only evolving one population of the two parties in cooperation for each generation, selecting the population to be evolved for the generation according to the cooperation strategy set by a balance mechanism and variants thereof, and determining the evolution content of the two parties in cooperation at present;
determining the current evolution situation according to the algorithm requirement and the current competitive situation of both parties: determining a population to be evolved according to the population reproduction rate and a weak population protection strategy for setting a threshold; according to a degeneration strong population strategy, when the inter-species difference is too large, the current generation is not normally evolved, but a strong population is selected to perform the evolution operation towards the degeneration direction, and the fitness value of an individual in the current strong population needs to be recalculated;
the competitive coevolution algorithm based on the equilibrium mechanism variant can still keep competition of two competing parties under the condition that the evolution search efficiency of the two competing parties is inconsistent or one party appears a strong individual, and avoids the phenomenon that the evolution is stuck to a halt, and specifically comprises the following steps:
s31: aiming at the obtained fitness values of the individuals of the two competing parties, reflecting the population strength of the two competing parties at present by defining a plurality of population strength measuring modes, and obtaining respective reproduction rates of the two competing parties according to the population strengths of the two competing parties, namely the probability that various populations are selected for evolution;
the population potential force intensity is calculated in two modes of match calculation and fitness mean calculation:
the first method is as follows: the competition calculation mode is that one individual is selected from the competition parties respectively, the appropriate value of the individual is compared with the appropriate value of the individual, the individual with the appropriate value wins, a certain number of competitions are carried out, and the proportion of the winning times of the competition parties in the total number of the competitions is obtained, namely the respective strength;
the second method comprises the following steps: the mode of calculating the mean value of the fitness values of the two competing populations is respectively calculated, and the strength of each population potential is the proportion of the mean value to the sum of the mean values
S32: aiming at the problems of inconsistent evolution search efficiency and large evolution speed difference of two competing parties, a balance mechanism is applied to solve the problems, namely, only one population is evolved in each generation in the specified evolution process, the probability that various populations are selected to be evolved, namely the reproduction rate, is determined by the population momentum strength and is inversely proportional to the population momentum strength, and the lower the population reproduction rate with stronger momentum, the smaller the probability of selected evolution;
wherein, the reproduction rate of various groups is inversely proportional to the strength of the potential force of the groups, and the strength of the potential force of the scene groups is assumed to be Str S The force intensity of the parameter population is 1-Str S Then the reproduction rate of the episodic population is 1-Str S The breeding rate of the parameter population is Str S ；
S33: judging whether the reproduction rate of the situation population reaches a threshold value according to a threshold value setting strategy which is set by an algorithm and aims at protecting the disadvantage population, and if the reproduction rate of the situation population does not reach the threshold value, directly judging to select the situation population for evolution;
s34: the method comprises the following steps of (1) setting a degeneration strong population strategy according to an algorithm and aiming at solving the problem that the situation that the other party population completely loses competitiveness due to the occurrence of strong individuals in the one party population possibly occurring in the evolution process: judging whether the potential force difference between the two populations is overlarge, and when the potential force difference between the two populations is overlarge, carrying out degeneration operation on the strong party, namely changing the fitness condition of the individuals, so that the individuals with backward ranking in the populations are selected to be evolved in the subsequent evolution selection process, thereby achieving the purpose of reducing the strength of the strong populations;
the degradation strong population strategy is characterized in that the fitness value of an individual in a strong population is recalculated, so that the individual which is originally in a low fitness value behind the ranking in the population obtains a higher fitness value, and is selected to participate in generation of filial generations with a higher probability, thereby reducing the overall strength of the strong population, balancing the two populations, keeping competition of two parties in competition and avoiding stagnation of evolution;
the fitness recalculation is as follows:
f′＝2ω 2 x-ωx 2
wherein f' is the recalculated individual fitness value, x is the score value of each individual in the interval of [0,1] obtained by normalizing the fitness values f of all the individuals of the dominant population, and omega is the degradation coefficient in the interval (0, 0.5) and represents the degradation degree of the population, and is determined by the difference between the populations, and the larger the difference is, the smaller the degradation coefficient is, the higher the degradation degree is;
the inferior population is a population with low evolution speed in the evolution process and is a situation population; the degeneration strong population is a population in a state of over-strong potential and complete defeat of adversaries;
s4: and (3) the two parties collaborate to evolve according to requirements: evolving the selected party of the cooperation according to the decision obtained in the step S3, wherein the situation population adopts an evolutionary method of genetic programming, and the parameter population adopts an evolutionary method of differential evolution;
when the situation population is selected to evolve or is used as a strong population to perform degradation operation, genetic programming evolution based on grammar guidance is adopted for the situation population coded by the grammar deduction tree; when selecting the parameter group, adopting a differential evolution method for the parameter group coded by the real number array;
when selecting the parameter population, adopting a differential evolution method for the parameter population coded by the real number array: the differential mutation operator is shown below:
wherein the content of the first and second substances,
is a new individual obtained by mutation operation of the ith individual of the G generation,
represents the ith individual of the G-th generation,
is any one of the top p% best individuals, r1 and r2 are any two individuals in the G generation population, and F is a scaling factor;
the differential crossover operator is shown as follows:
wherein the content of the first and second substances,
is a new individual obtained by the ith individual of the G generation through the crossing operation, Cr is the crossing rate, j is rand An integer defining a random position selected for each individual;
the difference algorithm applies a greedy selection strategy and hopes to retain the individuals with high fitness to the next generation, namely when the fitness of the new individual obtained after the variation crossing operation is larger than that of the old individual, the new individual is retained to the next generation, otherwise, the old individual is retained to the next generation, as shown in the following formula:
2. the evaluation method of the high-frequency ground wave radar target tracking algorithm based on collaborative scene evolution according to claim 1, wherein in step S2, a test scene is tested under a target tracking algorithm corresponding to a parameter individual, and a track fracture degree is selected as an evaluation index of a tracking condition: when the track fracture degree exceeds a certain range, the target tracking algorithm under the parameter setting cannot track the test scene successfully.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011422657.9A CN113033754B (en) | 2020-12-08 | 2020-12-08 | Evaluation method for high-frequency ground wave radar target tracking algorithm based on collaborative scene evolution |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011422657.9A CN113033754B (en) | 2020-12-08 | 2020-12-08 | Evaluation method for high-frequency ground wave radar target tracking algorithm based on collaborative scene evolution |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113033754A CN113033754A (en) | 2021-06-25 |
| CN113033754B true CN113033754B (en) | 2022-09-13 |
Family
ID=76459042
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011422657.9A Active CN113033754B (en) | 2020-12-08 | 2020-12-08 | Evaluation method for high-frequency ground wave radar target tracking algorithm based on collaborative scene evolution |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113033754B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114859296B (en) * | 2022-04-26 | 2023-04-14 | 哈尔滨工业大学 | High-frequency ground wave radar signal processing method with space scanning characteristic |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101740029A (en) * | 2009-12-16 | 2010-06-16 | 深圳大学 | Three-particle cooperative optimization method applied to vector quantization-based speaker recognition |
| CN106501801A (en) * | 2016-09-28 | 2017-03-15 | 哈尔滨工程大学 | A kind of bistatic MIMO radar tracking based on chaos Symbiotic evolution on multiple populations |
| CN109711096A (en) * | 2019-01-22 | 2019-05-03 | 中国海洋大学 | Ship Target motion modeling and emulation mode based on the syntax |
| CN109840916A (en) * | 2019-01-22 | 2019-06-04 | 中国海洋大学 | The assessment method of high-frequency ground wave radar Ship Trakding Algorithm |
| CN110097088A (en) * | 2019-04-08 | 2019-08-06 | 燕山大学 | A kind of dynamic multi-objective evolvement method based on transfer learning Yu particular point strategy |
| CN110134007A (en) * | 2019-05-22 | 2019-08-16 | 南昌航空大学 | Multiple no-manned plane cooperates with target assignment method |
| CN110598863A (en) * | 2019-09-05 | 2019-12-20 | 南宁师范大学 | Multi-target differential evolution method for co-evolution |
| CN111061995A (en) * | 2019-11-28 | 2020-04-24 | 江南机电设计研究所 | Combat resource allocation method, first equipment and second equipment |
-
2020
- 2020-12-08 CN CN202011422657.9A patent/CN113033754B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101740029A (en) * | 2009-12-16 | 2010-06-16 | 深圳大学 | Three-particle cooperative optimization method applied to vector quantization-based speaker recognition |
| CN106501801A (en) * | 2016-09-28 | 2017-03-15 | 哈尔滨工程大学 | A kind of bistatic MIMO radar tracking based on chaos Symbiotic evolution on multiple populations |
| CN109711096A (en) * | 2019-01-22 | 2019-05-03 | 中国海洋大学 | Ship Target motion modeling and emulation mode based on the syntax |
| CN109840916A (en) * | 2019-01-22 | 2019-06-04 | 中国海洋大学 | The assessment method of high-frequency ground wave radar Ship Trakding Algorithm |
| CN110097088A (en) * | 2019-04-08 | 2019-08-06 | 燕山大学 | A kind of dynamic multi-objective evolvement method based on transfer learning Yu particular point strategy |
| CN110134007A (en) * | 2019-05-22 | 2019-08-16 | 南昌航空大学 | Multiple no-manned plane cooperates with target assignment method |
| CN110598863A (en) * | 2019-09-05 | 2019-12-20 | 南宁师范大学 | Multi-target differential evolution method for co-evolution |
| CN111061995A (en) * | 2019-11-28 | 2020-04-24 | 江南机电设计研究所 | Combat resource allocation method, first equipment and second equipment |
Non-Patent Citations (3)
| Title |
|---|
| "A competitive co-evolutionary approach for the multi-objective evolutionary algorithms";VAN TRUONG VU 等;《IEEE》;20161231;第1-23页 * |
| "A Performance Evaluation Scheme for Multiple Object Tracking with HFSWR";Kun Wang 等;《MDPI》;20191231;第1-20页 * |
| "An Evolutionary Vulnerability Detection Method for HFSWR Ship Tracking Algorithm";Pengju Zhang 等;《Springer》;20171231;第763-773页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113033754A (en) | 2021-06-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108724182B (en) | End-to-end game robot generation method and system based on multi-class simulation learning | |
| Yen et al. | Two-stage Monte Carlo tree search for Connect6 | |
| JP2013081683A (en) | Information processing apparatus, information processing method, and program | |
| CN112491818B (en) | Power grid transmission line defense method based on multi-agent deep reinforcement learning | |
| CN113033754B (en) | Evaluation method for high-frequency ground wave radar target tracking algorithm based on collaborative scene evolution | |
| US20220374722A1 (en) | Intelligent ammunition co-evolution task assignment method | |
| Xiao et al. | Network security situation prediction method based on MEA-BP | |
| CN113708976B (en) | Heterogeneous combat network collapse method based on deep reinforcement learning | |
| Nakashima et al. | An evolutionary approach for strategy learning in robocup soccer | |
| CN111412795A (en) | Test point setting scheme generation method and device | |
| CN110852574A (en) | Target threat assessment method and medium based on improved grey target theory | |
| CN113689001B (en) | Virtual self-playing method and device based on counter-facts regretation minimization | |
| CN112755538B (en) | Real-time strategy game match method based on multiple intelligent agents | |
| CN113255883A (en) | Weight initialization method based on power law distribution | |
| CN112870727A (en) | Training and control method for intelligent agent in game | |
| CN110443344A (en) | A kind of momenttum wheel method for diagnosing faults and device based on K2ABC algorithm | |
| Liang et al. | A Study on the Agent in Fighting Games Based on Deep Reinforcement Learning | |
| CN114510876B (en) | Multi-platform weapon target distribution method based on symbiotic search biological geography optimization | |
| CN111481935B (en) | Configuration method, device, equipment and medium for AI models of games with different styles | |
| CN116510301A (en) | Game character behavior control method and device and electronic equipment | |
| Ontanon | Experiments on learning unit-action models from replay data from rts games | |
| Thawonmas et al. | Aggregation of action symbol sub-sequences for discovery of online-game player characteristics using KeyGraph | |
| CN112418384A (en) | Competitive coevolution algorithm for solving asymmetric tasks based on balance mechanism variants | |
| Nie et al. | Improve Robustness of Reinforcement Learning against Observation Perturbations via l∞ Lipschitz Policy Networks | |
| Chen et al. | Research on turn-based war chess game based on reinforcement learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |