A Multi-Objective Allocation Method for Heterogeneous Unmanned Vehicle Clusters

Abstract

This invention provides a multi-target allocation method for heterogeneous unmanned vehicle (UAV) swarms. The method includes generating the threat levels of targets (E) and heterogeneous UAVs (M). An individual assessment and dynamic fine-tuning module performs individual assessments and dynamic fine-tuning of UAVs and targets based on the threat levels of targets (E) and heterogeneous UAVs (M). If the individual assessment and dynamic fine-tuning reaches the maximum number of iterations, the optimal target allocation scheme is output. If the maximum number of iterations is not reached, a differential evolution module with a repair mechanism performs population individual repair processing on the UAVs and targets. After repair, step 3 is repeated until the optimal target allocation scheme is obtained. This invention can achieve target allocation even when the number of UAVs and targets is unequal. By introducing a population repair mechanism, it satisfies the capability constraints of heterogeneous UAVs. A dynamic ε-Nash equilibrium mechanism guides the population individuals to evolve towards a "high threat - high survival" direction.

Description

Technical Field

[0001] This invention relates to the field of cooperative control technology for unmanned systems, and in particular to a multi-objective allocation method for heterogeneous unmanned vehicle clusters. Background Technology

[0002] Currently, many target allocation methods for multi-autonomous vehicle swarms still rely on "1v1" or "homogeneous unmanned vehicles" as prerequisites, implicitly assuming two basic assumptions: first, the number of unmanned vehicles and targets is the same; second, all executors have homogeneous capabilities, and the threat probability to all targets can be simplified to satisfying the same probability distribution. However, in asymmetric, highly adversarial real-world scenarios, both assumptions are broken: the number of unmanned vehicles (M side) and targets (E side) is rarely exactly the same; the capabilities of vehicles in a heterogeneous unmanned vehicle swarm differ, allowing for combined advantages, thus the threat situation of different vehicles to the same target may also differ.

[0003] To address the aforementioned issues, there is an urgent need for a multi-objective allocation method for heterogeneous unmanned vehicle clusters, which can solve the problems existing in traditional methods. Summary of the Invention

[0004] The purpose of this invention is to provide a multi-target allocation method for heterogeneous unmanned vehicle clusters.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0006] A multi-objective allocation method for heterogeneous autonomous vehicle swarms includes:

[0007] Step 1: Generate the target threat level of Party E based on the target threat level generation module;

[0008] Step 2: Generate the M-square heterogeneous unmanned vehicle threat level based on the M-square heterogeneous unmanned vehicle threat level generation module;

[0009] Step 3: Based on the individual assessment and dynamic fine-tuning module, perform individual assessment and dynamic fine-tuning of unmanned vehicles and targets according to the threat level of E-side targets and the threat level of heterogeneous unmanned vehicles of M-side targets. If the individual assessment and dynamic fine-tuning reaches the maximum number of iterations, the optimal target allocation scheme is output.

[0010] Step 4: If the maximum number of iterations has not been reached, then the autonomous vehicle and the target are subjected to population individual repair processing based on the differential evolution module with repair mechanism. After the repair is completed, Step 3 is repeated until the optimal target allocation scheme is obtained.

[0011] Further, in step 1, the target threat level of party E is generated based on the target threat level generation module, specifically as follows:

[0012] The objectives of Party E are categorized into two types: known and unknown.

[0013] When the target is known, the threat probability of that target to the M-party unmanned vehicle satisfies (μ e , σ e The truncated normal distribution;

[0014] When the target is unknown, a threat assessment algorithm based on fuzzy theory evaluates the threat level of the unknown target.

[0015] Furthermore, the threat assessment algorithm based on fuzzy theory assesses the threat level of unknown targets, specifically as follows:

[0016] Using the target size and moving speed as inputs, fuzzification is performed using a trapezoidal membership function;

[0017] Calculate the rule activation strength based on a preset fuzzy rule base;

[0018] The threat level value is obtained by performing defuzzification using a weighted average method.

[0019] Furthermore, in step 2, the M-party heterogeneous autonomous vehicle threat level is generated based on the M-party heterogeneous autonomous vehicle threat level generation module, specifically as follows:

[0020] The heterogeneous unmanned vehicle clusters are classified into three categories: Class I, Class II, and Class III. Among them, Class I vehicles satisfy the truncated normal distribution of threat level (μ1, σ1), which has a low threat level to specific targets.

[0021] Class II vehicles satisfy a truncated normal distribution of threat level (μ2, σ2), indicating a low threat level to specific targets;

[0022] Class I vehicles satisfy a truncated normal distribution of threat level (μ3, σ3), with a threat level of 0 against distant targets.

[0023] Furthermore, in step 3, the individual assessment and dynamic fine-tuning module performs individual assessments and dynamic fine-tuning of the unmanned vehicles and targets based on the threat level of target E and the heterogeneous unmanned vehicle threat level of unmanned vehicles M. If the individual assessment and dynamic fine-tuning reaches the maximum number of iterations, the optimal target allocation scheme is output, specifically as follows:

[0024] After generating the threat level of target E and the threat level of heterogeneous unmanned vehicles M, the parameters are initialized to obtain the preset parameters;

[0025] After generating an initial population based on preset parameters, fitness assessments and strategy adjustments are performed on individuals within the population.

[0026] Furthermore, the preset parameters include the number of unmanned vehicles N. A Target quantity N T Number of iterations G, population size Ω, differential evolution mutation factor F, differential evolution crossover factor CR, lower limit of threat level to target E pmin Lower limit of survival probability S for unmanned vehicles min The mean and variance of the threat levels of three types of unmanned vehicles (μ1, σ1), (μ2, σ2), and (μ3, σ3), and the mean and variance of the threat levels of known targets (μ e , σ e ).

[0027] Furthermore, after generating an initial population based on preset parameters, fitness assessment and strategy adjustment are performed on individuals in the population, specifically as follows:

[0028] For each individual in the population, calculate the population individual fitness based on joint actions;

[0029] Based on the individual fitness of the population, each generation of individuals undergoes individual fine-tuning based on a dynamic ε-Nash equilibrium strategy.

[0030] Furthermore, in step 4, the autonomous vehicle and the target are subjected to population individual repair processing based on the differential evolution module with repair mechanism, specifically as follows:

[0031] The initial population was subjected to mutation treatment;

[0032] Cross-processing is performed on populations that are easier to handle;

[0033] Population restoration treatment is performed on the populations after cross-processing;

[0034] Assess the new population after population restoration treatment;

[0035] A selection process is performed on the new population after evaluation.

[0036] In summary, the present invention has at least one of the following beneficial technical effects:

[0037] 1. Currently, many target allocation methods for multi-vehicle clusters still rely on "1v1 confrontation" as a prerequisite, which requires the number of vehicles and targets to be equal. However, this invention uses a population repair mechanism to repair the uniqueness of target allocation and the full coverage of targets during the population evolution process. This ensures that targets are not repeatedly assigned to the same vehicle, and all targets are allocated. This effectively gets rid of the constraint that the number of vehicles and targets must be equal due to "1v1". For any number of vehicles and targets, this invention can successfully allocate targets, greatly improving the flexibility of scenario applications.

[0038] 2. Most autonomous vehicle (V2V) swarms require homogeneous V2Vs for target allocation. However, in real-world scenarios, heterogeneous V2Vs have different capabilities and can leverage combined advantages, so they cannot be treated equally. This invention considers the capability constraints of heterogeneous V2Vs from multiple perspectives to ensure effective threats to the allocated targets. On one hand, it assumes that the threat levels of the three types of V2Vs follow different truncated normal distributions and establishes a constraint matrix of V2Vs' threats to targets, thus constructing a heterogeneous V2V swarm. On the other hand, in the population repair mechanism, this invention repairs target allocation schemes that violate heterogeneous capability constraints. This invention can fully unleash the combined counter-attack potential of the V2V swarm's "multi-vehicle collaboration and complementary advantages."

[0039] 3. To address the problem that traditional evolutionary algorithms have limited global exploration capabilities, this invention adopts a dynamic ε-Nash equilibrium strategy in the individual fitness assessment stage of the population. Through phased relaxation of game stability judgment, the strict zero deviation condition of traditional Nash equilibrium is transformed into a tolerable payoff threshold. The three-stage dynamic ε design enables the population to achieve a dynamic balance between strategy stability and global exploration during evolution. Attached Figure Description

[0040] Figure 1 This is a schematic diagram of the method flow of the present invention;

[0041] Figure 2 This is a schematic diagram illustrating the principle of a threat assessment algorithm based on fuzzy theory.

[0042] Figure 3 A schematic diagram of a dynamic ε-Nash equilibrium strategy;

[0043] Figure 4 A schematic diagram illustrating the target allocation results for a heterogeneous unmanned vehicle cluster.

[0044] Figure 5a To optimize the progress curve diagram;

[0045] Figure 5b This is a schematic diagram of the ε-Nash equilibrium evolutionary process curve;

[0046] Figure 5c This is a schematic diagram of the dynamic ε adjustment strategy. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0048] The purpose of this invention is to provide a multi-objective allocation method for heterogeneous unmanned vehicle (UAV) swarms. By introducing game theory, the method comprehensively considers the threat level of each UAV and the target, as well as the survival probability of each UAV in the swarm, during the fitness evaluation of individuals in each generation. A dynamic ε-Nash equilibrium mechanism is used to adjust individual encoding, further identifying individuals with better fitness in the current offspring generation, guiding the population towards a "high threat - high survival" direction. In the differential evolution stage of each generation, this invention also introduces a population repair mechanism to ensure that individuals after crossover and mutation meet the capability constraints of heterogeneous UAVs. Ultimately, this achieves a reasonable allocation of multiple objectives by the heterogeneous UAV swarm, significantly improving the execution efficiency of the heterogeneous UAV swarm in multi-objective tasks.

[0049] like Figure 1 As shown, this invention provides a multi-target allocation method for heterogeneous unmanned vehicle clusters, including:

[0050] Step 1: Generate the target threat level of Party E based on the target threat level generation module;

[0051] Step 2: Generate the M-square heterogeneous unmanned vehicle threat level based on the M-square heterogeneous unmanned vehicle threat level generation module;

[0052] Step 3: Based on the individual assessment and dynamic fine-tuning module, perform individual assessment and dynamic fine-tuning of unmanned vehicles and targets according to the threat level of E-side targets and the threat level of heterogeneous unmanned vehicles of M-side targets. If the individual assessment and dynamic fine-tuning reaches the maximum number of iterations, the optimal target allocation scheme is output.

[0053] Step 4: If the maximum number of iterations has not been reached, then the autonomous vehicle and the target are subjected to population individual repair processing based on the differential evolution module with repair mechanism. After the repair is completed, Step 3 is repeated until the optimal target allocation scheme is obtained.

[0054] In step 1, the threat level of target E is generated based on the target threat level generation module, specifically as follows:

[0055] The objectives of Party E are categorized into two types: known and unknown.

[0056] When the target is known, the threat probability of that target to the M-party unmanned vehicle satisfies (μ e , σ e The truncated normal distribution;

[0057] When the target is unknown, a threat assessment algorithm based on fuzzy theory evaluates the threat level of the unknown target.

[0058] A threat assessment algorithm based on fuzzy theory evaluates the threat level of unknown targets, specifically as follows:

[0059] Using the target size and moving speed as inputs, fuzzification is performed using a trapezoidal membership function;

[0060] Calculate the rule activation strength based on a preset fuzzy rule base;

[0061] The threat level value is obtained by performing defuzzification using a weighted average method.

[0062] In step 2, the threat level of the heterogeneous autonomous vehicle is generated based on the M-square heterogeneous autonomous vehicle threat level generation module, specifically as follows:

[0063] The heterogeneous unmanned vehicle clusters are classified into three categories: Class I, Class II, and Class III. Among them, Class I vehicles satisfy the truncated normal distribution of threat level (μ1, σ1), which has a low threat level to specific targets.

[0064] Class II vehicles satisfy a truncated normal distribution of threat level (μ2, σ2), indicating a low threat level to specific targets;

[0065] Class I vehicles satisfy a truncated normal distribution of threat level (μ3, σ3), with a threat level of 0 against distant targets.

[0066] In step 3, the individual assessment and dynamic fine-tuning module performs individual assessments and dynamic fine-tuning of unmanned vehicles and targets based on the threat level of target E and the heterogeneous unmanned vehicle threat level of unmanned vehicles M. If the individual assessment and dynamic fine-tuning reaches the maximum number of iterations, the optimal target allocation scheme is output, specifically:

[0067] After generating the threat level of target E and the threat level of heterogeneous unmanned vehicles M, the parameters are initialized to obtain the preset parameters;

[0068] After generating an initial population based on preset parameters, fitness assessments and strategy adjustments are performed on individuals within the population.

[0069] The preset parameters include the number of autonomous vehicles N. A Target quantity N T Number of iterations G, population size Ω, differential evolution mutation factor F, differential evolution crossover factor CR, lower limit of threat level to target E p min Lower limit of survival probability S for unmanned vehicles min The mean and variance of the threat levels of three types of unmanned vehicles (μ1, σ1), (μ2, σ2), and (μ3, σ3), and the mean and variance of the threat levels of known targets (μ e , σ e ).

[0070] After generating an initial population based on preset parameters, the fitness of individuals in the population is assessed and strategies are adjusted, specifically as follows:

[0071] For each individual in the population, calculate the population individual fitness based on joint actions;

[0072] Based on the individual fitness of the population, each generation of individuals undergoes individual fine-tuning based on a dynamic ε-Nash equilibrium strategy.

[0073] In step 4, the individual population repair process for the unmanned vehicle and the target is performed based on the differential evolution module with a repair mechanism, specifically as follows:

[0074] The initial population was subjected to mutation treatment;

[0075] Cross-processing is performed on populations that are easier to handle;

[0076] Population restoration treatment is performed on the populations after cross-processing;

[0077] Assess the new population after population restoration treatment;

[0078] A selection process is performed on the new population after evaluation.

[0079] This invention provides a specific embodiment to elaborate on the above method, specifically as follows:

[0080] In practical applications, m autonomous vehicles perform preset tasks on n targets. To ensure a high success rate (i.e., a high threat level) when two autonomous vehicles perform tasks on the same target, it is assumed that the two vehicles can simultaneously perform preset actions on the target. If the joint action fails, the target will take countermeasures against the vehicles, and the vehicle with a higher threat level is more likely to be countered. The flowchart of this invention is as follows for the above scenario. Figure 1 As shown.

[0081] The implementation of this invention mainly consists of four parts: an E-side target threat generation module, an M-side heterogeneous unmanned vehicle threat generation module, an individual assessment and dynamic fine-tuning module, and a differential evolution module with a population repair mechanism.

[0082] 1. The E-side target threat generation module sets E-side targets into two categories: known and unknown. When a target is known, the threat probability of that target to M-side vehicles is assumed to follow a truncated normal distribution, as shown in the following formula:

[0083] (1)

[0084] In the formula, μ e σ is the mean. e Standard deviation It is a cutoff interval. In real-world scenarios, the threat capability of a target fluctuates normally. It is unreasonable for the threat probability to be too high or too low. Therefore, it is necessary to cut off the assumed normal distribution.

[0085] When the target is unknown, a threat assessment algorithm based on fuzzy theory is proposed. The principle is as follows: Figure 2As shown. In real-world scenarios, it is difficult to obtain information such as the payload type and payload volume of an unknown target (E), while the target size and speed are readily apparent. Generally, target size reflects E's payload carrying capacity, while speed reflects E's maneuverability and the weight of the potential payload. These two factors interact and jointly determine the current threat level of the target. Therefore, we use the target's size and speed as inputs and introduce fuzzy theory to assess the target's threat level. To simplify calculations, target length is used to represent size.

[0086] This invention uses a trapezoidal membership curve to fuzzify the target size and target movement speed. The formula required for fuzzification is as follows:

[0087] Formulas (2)-(4) define the membership function of the target size, which is...

[0088] (2)

[0089] (3)

[0090] (4)

[0091] In the formula, x represents the actual size of the target, in meters; , , These represent the membership values ​​of the target size as belonging to small, medium, and large sizes, respectively. For each actual target size, the membership value for belonging to small, medium, and large sizes is calculated. , , ;

[0092] Formulas (5)-(7) define the membership function of the target's moving speed, specifically:

[0093] (5)

[0094] (6)

[0095] (7)

[0096] Where v represents the target’s actual moving speed, in kilometers per hour; , , These represent the membership values ​​indicating whether a target's movement speed is slow, medium, or fast. For each type of actual target movement speed, the membership value indicating whether it belongs to slow, medium, or fast is calculated. , , .

[0097] After fuzzing the target size x and moving speed v using equations (2) to (7), fuzzy reasoning calculations are required based on the fuzzy rule table. Table 1 combines all possible target sizes and moving speeds after fuzzing, and gives the reference threat level and rule weights of different types of targets according to actual scenario experience. The fuzzy rule library is defined as shown in Table 1.

[0098] Table 1. Schematic diagram of fuzzy rule base

[0099]

[0100] The following is an explanation of Table 1. The reference threat level is derived based on experience in real-world scenarios, and the specific details are as follows:

[0101] For small-sized targets, limited by physical space, they can typically only carry lightweight weapons, resulting in weaker attack range and destructive power. At slow speeds, they are easily intercepted, leading to low attack effectiveness; at medium speeds, they possess some penetration capability; at high speeds, they are prone to instability in control, reducing their actual threat. Therefore, based on practical experience, the reference threat levels for R1, R2, and R3 are set to 0.65, 0.75, and 0.65, respectively.

[0102] For medium-sized targets of side E, more weapons can be carried, but changes in speed significantly affect their threat performance. At low, slow speeds, targets may carry high-powered weapons or ample ammunition, increasing the threat; therefore, the reference threat value for R4 is set at 0.9. At medium speeds, targets are usually in a ready-to-go state, and the threat is manageable; therefore, the reference threat value for R5 is set at 0.75. At high speeds, weapons are reduced in load, making them vulnerable to rapid destruction; therefore, the reference threat value for R6 is set to the lowest level of 0.65.

[0103] For large E-type targets, which possess strong attack potential (such as the possibility of integrating heavy systems), speed reveals their tactical intentions. At slow speeds, they are likely deploying high-lethality weapons and pose the greatest threat; therefore, the reference threat value for R7 is set to the highest at 0.95. At medium speeds, they still retain strong attack capabilities; therefore, the reference threat value for R8 is set to a relatively high 0.85. At fast speeds, it is highly likely that the target is retreating or relocating, and its attack intentions are weakened; therefore, the reference threat value for R9 is set to the lowest at 0.65.

[0104] The rule weights are allocated based on the credibility of the threat level settings mentioned above. The more stable the threat level value of different size-speed combinations, the higher their rule weight. For example, R4, R7, and R8 have relatively stable threat level values, while R3, R6, and R9 have relatively low threat level values, so their rule weights are all relatively high. The threat level values ​​of other size-speed combinations have lower credibility, so their planning weights are also lower.

[0105] By combining the membership values ​​of target size and target movement speed with the fuzzy rule base, the rule activation strength is calculated. The formula is as follows:

[0106] (8)

[0107] In the formula, Let be the rule weight for the i-th planning rule; This indicates that the target belongs to a set of membership values ​​of different sizes; This represents the set of membership values ​​for which the target belongs to different movement speeds.

[0108] Based on the activation intensity according to the rules R1-R9, the defuzzification operation defined in equation (9) is performed to obtain the final target threat value. for:

[0109] (9)

[0110] Among them, c i Let be the reference threat level of the i-th rule; Let represent the set of rules. As can be easily seen from the above formula, the weighted average method for defuzzification has the advantages of being computationally simple and efficient, and also taking into account the contributions of all activation rules.

[0111] 2. M-Square Heterogeneous Unmanned Vehicle Threat Generation Module. Assume the heterogeneous unmanned combat vehicle cluster consists of three types of unmanned vehicles, each with certain limitations, as shown below:

[0112] (10)

[0113] Where a and t represent the autonomous vehicle index and the target index, respectively. , , This represents 5 randomly selected targets. It's easy to see from the formula that the first type of autonomous vehicle has limited threat capability against the targets, while the second type... Limited target threat capability, third type of autonomous vehicle The target's threat capability is limited.

[0114] The threat level of the three types of unmanned vehicles to a target satisfies the following formula:

[0115] (11)

[0116] The three types of driverless vehicles respectively meet ( , ), ( , ), ( , The truncated normal distribution represents the varying mission capabilities of the three types of autonomous vehicles due to their different payloads. Therefore, the actual threat level of each autonomous vehicle is a random value that follows its own truncated normal distribution.

[0117] 3. Individual Assessment and Dynamic Fine-Tuning Module. After the E-side target and M-side heterogeneous unmanned vehicle threat levels are generated, parameter initialization is performed first, including the number of unmanned vehicles. N A Target quantity N T Number of iterations G Population size Ω Differential evolutionary mutation factor F Differential evolution crossover factor CR Lower limit of threat level to target E p min Lower limit of survival probability for autonomous vehicles S min Mean and variance of threat levels of three types of unmanned vehicles ( , ), ( , ), ( , ), the known threat level mean and variance of the target ( , After generating an initial population based on preset parameters, fitness assessments and strategy adjustments are performed on individuals within the population.

[0118] For each individual in the population, calculate the individual fitness based on joint actions, using the following formula:

[0119] (12)

[0120] Among them, v a Let λ be the value of the autonomous vehicle 'a'; λ be the weight of the penalty term. This is a penalty item; The penalty coefficient for M's unmanned vehicle failing to complete the task according to the preset probability; P represents the survival probability penalty coefficient for driverless vehicles of side M; d (t) represents the total threat level of all vehicles in the M-side unmanned vehicles that head towards the E-side target t to perform the mission; A collection of unmanned vehicles heading to target t to perform a mission; Let M be the threat level of unmanned vehicle a to target t; T a The target set assigned to driverless vehicle a; S represents the threat level of target t to autonomous vehicle a; a Let be the survival probability of driverless car a; Indicates individual fitness based on joint action; Indicates the target quantity; Indicates the number of driverless cars; This indicates the lower limit of the threat level to target E; This represents the lower bound of the survival probability of autonomous vehicles; This represents the total number of M-side unmanned vehicles that travel to target t on side E to perform the mission; This represents the set of targets assigned to driverless vehicle a; This indicates the threat level of unmanned vehicle a to target t.

[0121] As shown in formula (12), the comprehensive fitness function considers the probability of the unmanned vehicle swarm successfully executing the preset tasks at each target and the survival probability of the unmanned vehicles. At the same time, individuals that do not meet the requirements of the success probability of threatening the target and the survival probability of the unmanned vehicles are penalized. Suppose that two unmanned vehicles will take action against the target simultaneously in order to ensure a high joint threat level. If the task execution against the target fails, the target will take countermeasures based on the threat capability of the vehicles that came to perform the task. That is, the vehicle with a higher threat level to the target has a greater probability of being countered by the target.

[0122] Based on the comprehensive fitness function, each generation of individuals is adjusted using a dynamic ε-Nash equilibrium strategy. Figure 3 The algorithm diagram shown first adjusts the individuals in the population based on the individual fitness of individuals before and after adjustment, respectively, based on joint action. Then, the final population individuals of the current generation are obtained through a dynamic ε-Nash equilibrium strategy.

[0123] The dynamic ε-Nash equilibrium strategy transforms the strict zero-deviation condition of traditional Nash equilibrium into a tolerable payoff threshold through a phased relaxation of game stability criteria. This achieves a dynamic balance between strategy stability and global exploration during evolution, as shown in the following formula:

[0124] (13)

[0125] Where I represents a single individual in the current population, i.e., a target allocation scheme; I' represents the adjusted current individual; I new This indicates that only strategies that improve fitness more than a threshold will be adopted; ε is the dynamic tolerance threshold.

[0126] ε shrinks with each generation, thus achieving an optimal balance between exploration and convergence in population evolution, overcoming the "early locking" defect of traditional methods. The threshold shrinkage expression is as follows:

[0127] (14)

[0128] Where g represents the current generation of the population, and G represents the total number of generations. As shown in (14), we divide the population evolution into three stages. In the early stage of evolution (the first 30% of generations)... This ensures that individuals in the population have ample room for improvement, preventing the population from converging too early and getting trapped in local optima; this occurs in the intermediate stages of evolution (30%–70% of generations). This can effectively improve the adjustment efficiency of individuals in a population; in the final stage of evolution (the last 30 generations). This causes the population to converge to a strictly approximate Nash equilibrium, thus yielding the final objective allocation scheme.

[0129] 4. Differential Evolution Module with Population Repair Mechanism. The differential evolution algorithm is a stochastic heuristic search algorithm that simulates the evolutionary development of biological populations in nature based on the principle of "survival of the fittest." It retains the global search strategy based on the population while introducing a one-to-one competitive survival strategy, reducing the complexity of genetic operations. Its main steps are as follows:

[0130] (1) The mutation operation of the differential evolution algorithm is shown in formula (15). This operation generates a differential vector with strong directionality and adjustable amplitude in the search space, effectively maintaining population diversity and avoiding premature convergence, providing the algorithm with global exploration capability, and enabling it to...

[0131] (15)

[0132] in This represents the mutation vector of the i-th individual in the g-th generation of the population; , , This represents three randomly selected parent individuals; It is a variable factor.

[0133] (2) The crossover operation of the differential evolution algorithm is shown in formula (16). This mechanism realizes the orderly recombination of parental and mutation information, introduces new information while retaining superior genes, and balances the trade-off between development and utilization, thus providing...

[0134] (16)

[0135] Cross factor , A random number between 0 and 1.

[0136] (3) The selection operation of the differential evolution algorithm is shown in formula (17). Through fitness comparison, individuals with high fitness are retained, and individuals with low fitness are replaced with individuals with high fitness mutations. This ensures that the population always evolves in the correct direction and avoids population degeneration.

[0137] (17)

[0138] To avoid illegal solutions generated by differential evolution under multiple constraints such as heterogeneous vehicle models and target coverage, this invention designs a population repair mechanism after each generation of evolutionary selection operations to repair individual populations and correct abnormal individuals generated by differential evolution. The abnormalities of individual populations are divided into four types: target number exceeding the limit, violation of heterogeneous capability constraints, duplicate target allocation, and incomplete target coverage. First, it checks whether the individual has the abnormality of target number exceeding the limit, not satisfying the heterogeneous capability constraint of equation (10), or duplicate target allocation. If so, it performs data range repair, heterogeneous capability constraint repair, and target allocation uniqueness repair respectively. Next, it checks whether the currently repaired individual has the abnormality of incomplete target coverage. If so, it performs target full coverage repair. Finally, the repaired individual population is obtained. For numerical range repair, the index of the target exceeds the target index is set to 0; for heterogeneous capability repair, the target index that does not meet the heterogeneous capability of the unmanned vehicle is set to 0; for target allocation uniqueness repair, the target index that is repeatedly allocated to the same unmanned vehicle is set to 0; for target full coverage repair, the unallocated target is randomly allocated to an unmanned vehicle that meets the capability constraints.

[0139] This invention was simulated and verified using MATLAB, with the following initial parameters: N A =8、N T =15, G=200, Ω=500, F=0.75, CR=0.8, p min =0.9, S min =0.7、 =0.8、 =0.052、 =0.85、 =0.05 2 , =0.9、 =0.05 2 .

[0140] Figure 4 The diagram shows the allocation results of the heterogeneous autonomous vehicle swarm, where the target threat probability is 0.93 and the autonomous vehicle survival rate is 0.91. Nearly 30% of the targets were successfully threatened by the autonomous vehicle swarm with a probability exceeding 95%, and all targets were successfully threatened with a probability exceeding 90%. All autonomous vehicles achieved a survival probability of over 70%.

[0141] Figure 5a -c gives the curve of the proposed differential evolution process. Figure 5a To optimize the progress curve, by Figure 5aIt can be seen that the target's threat level has reached its optimal fitness level by the 130th generation of the population. The M-type unmanned vehicle has a success rate of over 90% in performing tasks on the target and a survival rate of over 70%. Figure 5b The curve represents the evolutionary process of ε-Nash equilibrium. Figure 5c To dynamically adjust the ε-adjustment strategy curve, by Figure 5b and Figure 5c It can be seen that in about 20 generations of evolution across the three stages, the population tends to reach a near-Nash equilibrium, which makes it impossible for individuals to make effective fine adjustments. The dynamics of the ε value effectively breaks this deadlock.

[0142] Embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0143] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0144] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0145] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1The steps of the function specified in one or more boxes.

[0146] Contents not described in detail in this specification are prior art known to those skilled in the art. It is hereby indicated that the above description is intended to help those skilled in the art understand this invention, but does not limit the scope of protection of this invention. Any equivalent substitutions, modifications, improvements, or simplifications of the above descriptions that do not depart from the essential content of this invention fall within the scope of protection of this invention.

Claims

1. A multi-objective allocation method for heterogeneous unmanned vehicle clusters, characterized in that, include: Step 1: Classify the targets of Party E. Based on the target category, generate the threat level of Party E using the target threat level generation module, employing either a truncated normal distribution or a threat level assessment algorithm based on fuzzy theory. Step 2: Classify heterogeneous autonomous vehicle clusters, and generate M-square heterogeneous autonomous vehicle threat levels based on the M-square heterogeneous autonomous vehicle threat level generation module for different heterogeneous autonomous vehicle cluster categories; Step 3: Based on the individual assessment and dynamic fine-tuning module, parameters are initialized according to the target threat level of party E and the heterogeneous unmanned vehicle threat level of party M to obtain preset parameters. After generating an initial population based on the preset parameters, fitness assessment and strategy adjustment are performed on the individuals in the population. If the individual assessment and dynamic fine-tuning reaches the maximum number of iterations, the optimal target allocation scheme is output. For each individual in the population, the individual fitness based on joint action is calculated, as follows: （12） in, v a For driverless cars a Value; λ Weights for penalty items; This is a penalty item; The penalty coefficient for M's unmanned vehicle failing to complete the task according to the preset probability; The survival probability penalty coefficient for M-type unmanned vehicles; P d ( t ) refers to all unmanned vehicles from side M heading towards target side E. t The overall threat level of the vehicles performing the mission; To reach the target t A collection of driverless vehicles performing a mission; For M-type unmanned vehicles a For the target t The level of threat; T a For driverless cars a The target set to be allocated; For the goal t For driverless cars a The level of threat; S a For driverless cars a The probability of survival; Indicates individual fitness based on joint action; Indicates the target quantity; Indicates the number of driverless cars; This indicates the lower limit of the threat level to target E; This represents the lower bound of the survival probability of autonomous vehicles; Indicates driverless car a The target set to be allocated; Indicates M-side unmanned vehicle b For the target t The level of threat; Based on the comprehensive fitness function, individuals in each generation are adjusted using a dynamic ε-Nash equilibrium strategy. First, the population is adjusted, and the fitness of individuals before and after adjustment based on joint actions is calculated. Then, a dynamic ε-Nash equilibrium strategy is employed. ε The Nash equilibrium strategy yields the final population individuals in the current generation; dynamic ε The Nash equilibrium strategy achieves a dynamic balance between strategy stability and global exploration during evolution through phased relaxation of game stability criteria, as shown in the following formula: （13） in I This represents a single individual in the current population, i.e., a target allocation scheme; This represents the current individual after adjustment; This indicates that only strategies that improve fitness by more than a threshold will be adopted. ε This is a dynamic tolerance threshold; ε As the number of generations shrinks, the population evolution achieves an optimal balance between exploration and convergence. The threshold shrinkage expression is as follows: （14） in, g Indicates the current generation number of the population. G Indicates the total number of generations; Step 4: If the maximum number of iterations has not been reached, perform population individual repair processing on the unmanned vehicle and the target based on the differential evolution module with repair mechanism. After the repair is completed, repeat step 3 until the optimal target allocation scheme is obtained, specifically: The initial population was subjected to mutation treatment; Cross-processing is performed on populations that are easier to handle; Population restoration treatment is performed on the populations after cross-processing; Assess the new population after population restoration treatment; A selection process is performed on the new population after evaluation.

2. The method according to claim 1, characterized in that, In step 1, the threat level of target E is generated based on the target threat level generation module, specifically as follows: The objectives of Party E are categorized into two types: known and unknown. When the target is known, the threat probability of that target to the M-party unmanned vehicle satisfies (μ e , σ e The truncated normal distribution; When the target is unknown, a threat assessment algorithm based on fuzzy theory evaluates the threat level of the unknown target.

3. The method according to claim 2, characterized in that, A threat assessment algorithm based on fuzzy theory evaluates the threat level of unknown targets, specifically as follows: Using the target size and moving speed as inputs, fuzzification is performed using a trapezoidal membership function; Calculate the rule activation strength based on a preset fuzzy rule base; The threat level value is obtained by performing defuzzification using a weighted average method.

4. The method according to claim 3, characterized in that, In step 2, the threat level of the heterogeneous autonomous vehicle is generated based on the M-square heterogeneous autonomous vehicle threat level generation module, specifically as follows: The heterogeneous unmanned vehicle clusters are classified into three categories: Class I, Class II, and Class III. Among them, Class I vehicles satisfy the truncated normal distribution of threat level (μ1, σ1), which has a low threat level to specific targets. Class II vehicles satisfy a truncated normal distribution of threat level (μ2, σ2), indicating a low threat level to specific targets; Class I vehicles satisfy a truncated normal distribution of threat level (μ3, σ3), with a threat level of 0 against distant targets.

5. The method according to claim 4, characterized in that, The preset parameters include the number of autonomous vehicles N. A Target quantity N T Number of iterations G, population size Ω, differential evolution mutation factor F, differential evolution crossover factor CR, lower limit of threat level to target E p min Lower limit of survival probability S for unmanned vehicles min The mean and standard deviation of the threat levels of three types of unmanned vehicles (μ1, σ1), (μ2, σ2), and (μ3, σ3), and the mean and standard deviation of the threat levels of known targets (μ e , σ e ).

6. The method according to claim 5, characterized in that, After generating an initial population based on preset parameters, the fitness of individuals in the population is assessed and strategies are adjusted, specifically as follows: For each individual in the population, calculate the population individual fitness based on joint actions; Based on the individual fitness of the population, each generation of individuals undergoes individual fine-tuning based on a dynamic ε-Nash equilibrium strategy.

Images