A method and apparatus for optimizing multi-sensor task allocation and scheduling in unmanned swarms

Abstract

This invention provides a method and apparatus for optimizing the allocation and scheduling of multi-sensor tasks in unmanned swarms. It acquires unmanned swarm information and task information, encodes this information, and generates an initial population based on a probability distribution model using the encoded data. This initial population is then input into an iteration process. In each iteration, the population is optimized using fast non-dominated sorting, population mutation, and variable neighborhood local search to obtain an optimized population. The optimized population from each iteration is then transferred to an external archive set. Before each iteration, the probability distribution model is updated using the external archive set to improve the quality of the generated input population. This process is repeated until the external archive set yields an optimal solution that meets the given conditions. This solves the sensor task allocation problem in unmanned swarms and achieves target optimization in both time and resource consumption dimensions.

Description

Technical Field

[0001] This invention relates to the field of distributed cluster task planning, and in particular to a method and apparatus for optimizing the allocation and scheduling of multi-sensor tasks in unmanned clusters. Background Technology

[0002] In existing technologies, the dynamic and complex requirements for tasks achievable by unmanned swarms are increasing. Single sensors are insufficient to meet the needs of swarms, necessitating multi-sensor collaboration. Different sensors have varying functions and detection ranges, suitable for different tasks. Sensor resources are limited, the number of tasks that can be executed is also limited, and there is overlap in the functions and detection ranges of different sensors. A task can be completed by multiple sensors; therefore, an effective task allocation algorithm is needed. This algorithm enables multi-sensor collaboration by dynamically allocating tasks based on environmental changes. Furthermore, controlling the duration and resource consumption in the scheduling scheme is a crucial optimization objective, thereby further improving task completion benefits. Summary of the Invention

[0003] The technical problem to be solved by this invention is how to dynamically allocate tasks to unmanned swarms and optimize the allocation scheme in terms of duration and resource consumption.

[0004] The embodiments of the present invention adopt the following technical solutions:

[0005] Firstly, a method for optimizing the allocation and scheduling of multi-sensor tasks in unmanned swarms is provided, including:

[0006] Acquire unmanned cluster information and task information, encode the unmanned cluster information and task information, and generate an initial population through a probability distribution model and the encoded information;

[0007] The initial population is used as the input population and fed into the first iteration.

[0008] Each iteration includes: performing fast non-dominated sorting on the input population in terms of duration and resource consumption, and selecting high-quality individuals from the fast non-dominated sort as the first optimized population; mutating the first optimized population to generate a offspring population, merging the offspring population with the first optimized population for optimization, and obtaining the second optimized population; performing variable neighborhood local search optimization on each individual in the second optimized population in terms of duration and resource consumption, and obtaining the third optimized population; transferring the third optimized population to an external archive set, performing fast non-dominated sorting on all individuals in the external archive set to obtain an output individual, and determining whether the output individual meets the termination condition;

[0009] When the termination condition is not met, the probability distribution model is updated according to the external archive set. The updated probability distribution model generates a new input population and is input into the next iteration.

[0010] When the termination condition is met, the output individual is decoded to obtain the optimal solution.

[0011] Preferably, the step of acquiring unmanned swarm information and task information, and encoding the unmanned swarm information and task information, specifically includes:

[0012] Based on the unmanned cluster information, obtain information about all unmanned platforms; based on the information about all unmanned platforms, obtain information about each sensor in each unmanned platform; based on the task information, obtain information about all main tasks; based on the information about all main tasks, obtain information about each subtask in each main task.

[0013] The unmanned platform code is obtained based on the information of all unmanned platforms; the sensor code is obtained based on the information of all sensors; the main task code is obtained based on the information of all main tasks; and the subtask code is obtained based on the information of all subtasks.

[0014] Preferably, the step of generating the initial population through the probability distribution model and the encoded information specifically includes:

[0015] The correspondence between the main task code and the unmanned platform code in each individual is generated according to the probability distribution model, and the correspondence between the sub-task code and the sensor code in each individual is generated according to the probability distribution model, thereby generating a single individual, and multiple randomly generated individuals are used as the initial population;

[0016] The initial population includes: individuals of a first preset proportion generated randomly, individuals of a second preset proportion generated using a resource consumption strategy, and individuals of a third preset proportion generated using a duration strategy.

[0017] Preferably, the resource consumption strategy specifically includes:

[0018] When generating the correspondence between subtask codes and sensor codes, among all the sensors available for the subtask, the sensor code of the sensor with the least resource consumption is matched with the subtask code first, thereby obtaining the second preset ratio of individuals.

[0019] Preferably, the duration strategy specifically includes:

[0020] When generating the correspondence between subtask codes and sensor codes, among all the sensors available for the subtask, the sensor that has been occupied by other subtasks for the least amount of time is selected as the first type of sensor.

[0021] When the number of the first type of sensor is at most one, the sensor code and subtask code of the first type of sensor are matched to obtain the third preset ratio of individuals;

[0022] When the number of the first type of sensors is at least two, the sensor code and subtask code of the sensor with the shortest consumption time are matched first among the first type of sensors to obtain the third preset ratio of individuals.

[0023] Preferably, the step of performing a fast non-dominated sort on the input population along the time and resource consumption dimensions, and selecting high-quality individuals from the fast non-dominated sort as the first optimized population, specifically includes:

[0024] Obtain the dominance relationship between each individual in the input population and other individuals in terms of duration and resource consumption. Divide all individuals in the input population into levels according to the number of other individuals they dominate, and calculate the crowding degree for each level. Select the high-quality individuals based on the level division and crowding degree calculation.

[0025] Preferably, the step of performing variable neighborhood local search optimization on each individual in the second optimized population in terms of duration and resource consumption to obtain the third optimized population specifically includes:

[0026] The main task that occupies the most total sensor time and consumes the most total resources is selected as the first main task. The subtask with the latest completion time among all subtasks of the first main task is selected as the first subtask. The first subtask is executed first.

[0027] Select the unmanned platform that consumes the most total resources, and take the subtask with the most resource consumption in the main task corresponding to the unmanned platform as the second subtask. Obtain all sensors in the unmanned platform that can execute the second subtask, and associate the sensor with the least resource consumption in executing the second subtask with the second subtask.

[0028] Randomly select one or more unmanned platforms, randomly select a fourth preset proportion of sub-tasks from the main tasks corresponding to the unmanned platforms, reselect executable sensors for all the sub-tasks of the fourth preset proportion, and match the codes accordingly.

[0029] Preferably, the termination condition specifically includes:

[0030] Set the preset number of iterations:

[0031] Determine if the current iteration number is greater than or equal to the preset iteration number;

[0032] The termination condition is met when the current iteration number is greater than or equal to the preset iteration number.

[0033] The termination condition is not met when the current iteration number is less than the preset number of iterations.

[0034] Preferably, updating the probability distribution model based on the external archive set specifically includes:

[0035] The update formula for the corresponding probability between the main task code and the unmanned platform code is as follows:

[0036]

[0037] The update formula for the corresponding probability between subtask codes and sensors is as follows:

[0038]

[0039] Among them, the JF i,j (g+1) represents the probability corresponding to the main task code i and the unmanned platform code j in the (g+1)th iteration; the JF i,j (g) represents the probability corresponding to the main task code i and the unmanned platform code j in the g-th iteration; the PA c,d (g+1) represents the probability corresponding to the subtask code c and the sensor code d in the (g+1)th iteration; the PA c,d (g) represents the probability corresponding to subtask code c and sensor code d in the g-th iteration; α is the first preset learning rate, α∈(0,1); β is the second preset learning rate, β∈(0,1); Bestsize is the size of the external archive set;

[0040] The For the external archive set, according to the indicator function of the Z-th elite individual in the fast non-dominated sorting, when the main task code i and the unmanned platform code j correspond in the g-th iteration, When the main task code i and the unmanned platform code j do not correspond in the g-th iteration,

[0041] The For the external archive set, the indicative function of the Z-th elite individual in the fast non-dominated sorting is given when the subtask code c and the sensor code d correspond in the g-th iteration. When the subtask code c and the sensor code d do not correspond in the g-th iteration...

[0042] Secondly, an unmanned swarm multi-sensor task allocation and scheduling optimization device is provided, comprising at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the processor to perform the unmanned swarm multi-sensor task allocation and scheduling optimization method.

[0043] This invention provides a method and apparatus for optimizing the allocation and scheduling of multi-sensor tasks in unmanned swarms. It acquires unmanned swarm information and task information, encodes this information, and generates an initial population based on a probability distribution model using the encoded data. This initial population is then input into an iteration process. In each iteration, the population is optimized using fast non-dominated sorting, population mutation, and variable neighborhood local search to obtain an optimized population. The optimized population from each iteration is then transferred to an external archive set. Before each iteration, the probability distribution model is updated using the external archive set to improve the quality of the generated input population. This process is repeated until the external archive set yields an optimal solution that meets the given conditions. This solves the sensor task allocation problem in unmanned swarms and achieves target optimization in both time and resource consumption dimensions. Attached Figure Description

[0044] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments of the present invention will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0045] Figure 1 This is a flowchart of a method for optimizing the allocation and scheduling of multi-sensor tasks in an unmanned swarm provided by an embodiment of the present invention;

[0046] Figure 2 This is a schematic diagram of the encoding of an unmanned cluster multi-sensor task allocation and scheduling optimization method provided in an embodiment of the present invention;

[0047] Figure 3 This is a flowchart of the probability matrix normalization process for an unmanned cluster multi-sensor task allocation and scheduling optimization method provided in this embodiment of the invention.

[0048] Figure 4 This is a flowchart of a pre-optimization method in the time dimension of an unmanned cluster multi-sensor task allocation and scheduling optimization method provided in an embodiment of the present invention;

[0049] Figure 5 This is a schematic diagram of a variation of the unmanned cluster multi-sensor task allocation and scheduling optimization method provided in an embodiment of the present invention;

[0050] Figure 6 This is a schematic diagram corresponding to the encoding of an unmanned swarm multi-sensor task allocation and scheduling optimization method provided in an embodiment of the present invention;

[0051] Figure 7 This is a schematic diagram of an unmanned cluster multi-sensor task allocation and scheduling optimization device provided in an embodiment of the present invention. Detailed Implementation

[0052] 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.

[0053] In the description of this invention, the terms "inner", "outer", "longitudinal", "lateral", "upper", "lower", "top", "bottom", etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and do not require that this invention must be constructed and operated in a specific orientation. Therefore, they should not be construed as limiting this invention.

[0054] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0055] Example 1:

[0056] Embodiment 1 of the present invention provides an optimization method for multi-sensor task allocation and scheduling in unmanned swarms, such as... Figure 1 As shown, the method steps include:

[0057] In step 101, unmanned cluster information and task information are obtained, encoded according to the unmanned cluster information and task information, and an initial population is generated through a probability distribution model and the encoded information.

[0058] In this embodiment, the unmanned swarm includes multiple unmanned platforms, each of which includes multiple sensors. The unmanned swarm information includes information about each unmanned platform and information about each sensor within each unmanned platform. The task includes multiple main tasks, each of which includes multiple sub-tasks. The task information includes information about each main task and information about each sensor. It should be noted that in this embodiment, a single main task corresponds to a single unmanned platform, meaning that a main task is fully executed by a single unmanned platform, and each sub-task within a main task is executed by a single sensor within the unmanned platform corresponding to the main task. In other words, there is a one-to-one correspondence between the main task and the unmanned platform, and a one-to-one correspondence between the sub-task and the sensor.

[0059] The probability distribution model is a sampling probability matrix. In this embodiment, at least two probability matrices are defined. One probability matrix includes the corresponding probability of each main task and each unmanned platform, and the other probability matrix includes the corresponding probability of each sub-task and each sensor. Different corresponding situations are randomly obtained based on the above two probability matrices, thereby generating multiple individuals.

[0060] Based on the above coding rules, the probability distribution model generates multiple individuals. Each individual includes the main task corresponding to the unmanned platform, the sub-task corresponding to the sensor, the execution order of the sensor, the time taken for the sensor to complete the sub-task, and the resources required. The set of multiple individuals is the initial population.

[0061] In step 102, the initial population is input as the input population into the first round of iteration.

[0062] In step 103, each iteration includes: performing fast non-dominated sorting on the input population in terms of duration and resource consumption, and selecting high-quality individuals from the fast non-dominated sort as the first optimized population; mutating the first optimized population to generate a offspring population, merging the offspring population with the first optimized population for optimization, and obtaining the second optimized population; performing variable neighborhood local search optimization on each individual in the second optimized population in terms of duration and resource consumption, and obtaining the third optimized population; transferring the third optimized population to an external archive set, and performing fast non-dominated sorting on all individuals in the external archive set to obtain the output individual.

[0063] In step 104, it is determined whether the output individual meets the termination condition.

[0064] The step of performing fast non-dominated sorting on the input population in terms of duration and resource consumption is to compare and sort all individuals in the input population with the goal of shorter duration and less resource consumption. However, since it is difficult to simultaneously consider both duration and resource consumption, fast non-dominated sorting is used to customize the degree of consideration for each of duration and resource consumption. All individuals in the input population are sorted according to this definition, and the top preset number of individuals in the sort are the high-quality individuals. The high-quality individuals are extracted as the first optimized population.

[0065] The variable neighborhood local search optimization is to change the individuals in the second optimization population according to predetermined rules, and optimize towards the optimal solution during the change process. In this embodiment, the optimization direction for the allocation and scheduling of tasks and multi-sensor unmanned swarms is to minimize the time and resources required to execute the task.

[0066] The external archive set is the set of output populations in each iteration. When there is an individual in the external archive set that can satisfy the termination condition, it means that the individual is the optimal solution.

[0067] In step 105, when the termination condition is not met, the probability distribution model is updated according to the external archive set, and a new output population is generated according to the updated probability distribution model and input into the next iteration.

[0068] In step 106, when the termination condition is met, the output individual is decoded to obtain the optimal solution.

[0069] The decoding refers to decoding a single individual. This decoding operation is required not only when the optimal individual is finally obtained, but also during the iteration process when performing fast non-dominated sorting and variable neighborhood local search, so as to obtain the duration and resource consumption of the scheduling scheme represented by each individual.

[0070] In this embodiment, since the correspondence between the main task and the unmanned platform, as well as the correspondence between the sub-task and the sensor, needs to be encoded separately, and then generated and optimized separately according to different probabilities in the subsequent process, this embodiment also involves the following design:

[0071] Acquire unmanned cluster information and task information, and encode the unmanned cluster information and task information accordingly, specifically including:

[0072] Based on the unmanned cluster information, obtain information about all unmanned platforms; based on the information about all unmanned platforms, obtain information about each sensor in each unmanned platform; based on the task information, obtain information about all main tasks; based on the information about all main tasks, obtain information about each subtask in each main task.

[0073] The unmanned platform code is obtained based on the information of all unmanned platforms; the sensor code is obtained based on the information of all sensors; the main task code is obtained based on the information of all main tasks; and the subtask code is obtained based on the information of all subtasks.

[0074] Based on the two correspondences, two encoding schemes are set, such as... Figure 2 As shown, one type is the PV encoding for the main task and the unmanned platform. r ,For example Figure 2 In the middle pv1:[2,3], it refers to the main task coded as 2 corresponding to the unmanned platform coded as 3, that is, the main task coded as 2 is executed on the unmanned platform coded as 3; another type is the encoding jv for subtasks and sensors. r ,For example Figure 2The upper encoding string of encoding jv2 is the sensor encoding sequence, and the lower encoding string of encoding jv2 is the subtask encoding sequence. The sensor encoding and subtask encoding located in the same column of encoding jv2 are in a corresponding relationship.

[0075] In this embodiment, the probability distribution model is a sampling probability matrix. At least two probability matrices are defined: one includes the corresponding probabilities of each main task and each unmanned platform, and the other includes the corresponding probabilities of each sub-task and each sensor. Different correspondences are randomly obtained based on these two probability matrices, thereby generating multiple individuals. Therefore, this embodiment also involves the following design:

[0076] The generation of the initial population through the probability distribution model and the encoded information specifically includes:

[0077] The probability distribution model generates a correspondence between the main task code and the unmanned platform code in each individual, and generates a correspondence between the sub-task code and the sensor code in each individual, thereby generating a single individual. Multiple randomly generated individuals serve as the initial population.

[0078] In this embodiment, the probability distribution model includes two probability matrices, one of which, JF(g), represents the probability of the correspondence between each main task and all unmanned platforms, as follows:

[0079]

[0080]

[0081] Among them, the JF i,j (g) represents the probability of the main task encoded i and the unmanned platform encoded j corresponding in the g-th iteration, where n is the total number of main tasks, f is the total number of unmanned platforms, and JF... i,j (0) is This means that in the initial population generated in the first iteration, in order to ensure the dispersion of the population, the probability distribution model has a uniform distribution for the corresponding probabilities of the main task and the unmanned platform, that is, the corresponding probability of each main task for each unmanned platform is the same.

[0082] Another probability matrix PA(g) represents the probability of the correspondence between each subtask and all sensors in the corresponding unmanned platform, as follows:

[0083]

[0084]

[0085] Among them, PA c,d(g) represents the probability of the subtask encoding c and the sensor encoding d corresponding in the g-th iteration, where q is the total number of subtasks, a is the total number of sensors, and PA is the probability of the subtask encoding c and the sensor encoding d. c,f (0) is This means that in the initial population generated in the first iteration, in order to ensure the dispersion of the population, the probability distribution model has a uniform distribution for the corresponding probabilities of subtasks and sensors, that is, the corresponding probability of each subtask for each sensor is the same.

[0086] Within the algorithmic framework, individuals in the input population are generated using the two probability matrices mentioned above. The method for randomly generating the correspondences within these probability matrices is a roulette wheel strategy, such as... Figure 3 As shown, the method is as follows:

[0087] In step 201, each row of the probability matrix JF(g) is normalized, as shown in the following formula:

[0088]

[0089] Among them, the prob_JF i,j (g) represents the selection probability that the main task of code i is assigned to the unmanned platform of code j in the g-th iteration, where s is the code of any unmanned platform.

[0090] In step 202, a probability value δ∈(0,1) is randomly generated.

[0091] In step 203, when δ≤prob_JF i,1 (g) If so, the main task of code i is assigned to the unmanned platform of code 1; otherwise, proceed to step 204.

[0092] In step 204, when prob_JF i,j-1 (g)≤δ≤prob_JF i,j When (g) occurs, the main task of code i is assigned to the unmanned platform of code j.

[0093] In this embodiment, the correspondence among each individual is randomly generated through the probability distribution model. Furthermore, to further accelerate the subsequent optimization and reduce the number of iterations, this embodiment pre-optimizes the initial population when generating it, thereby greatly accelerating the subsequent optimization process. Therefore, this embodiment also involves the following design:

[0094] The initial population includes: individuals of a first preset proportion generated randomly, individuals of a second preset proportion generated using a resource consumption strategy, and individuals of a second preset proportion generated using a duration strategy.

[0095] The resource consumption strategy specifically includes:

[0096] When generating the correspondence between subtask codes and sensor codes, among all the sensors available for the subtask, the sensor code of the sensor with the least resource consumption is matched with the subtask code first, thereby obtaining the second preset ratio of individuals.

[0097] The duration strategy specifically includes, for example: Figure 4 As shown, the method flow is as follows:

[0098] In step 301, when generating the correspondence between subtask codes and sensor codes, among all the sensors available for the subtask, the sensor that has been occupied by other subtasks for the least amount of time is selected as the first type of sensor.

[0099] In step 302, when the number of the first type of sensor is at most one, the sensor code and subtask code of the first type of sensor are matched to obtain the third preset ratio of individuals.

[0100] In step 303, when the number of the first type of sensors is at least two, the sensor code and subtask code of the sensor with the shortest consumption time are matched first among the first type of sensors to obtain the third preset ratio of individuals.

[0101] In this embodiment, the first preset ratio, the second preset ratio, and the second preset ratio are all set by those skilled in the art according to the actual situation.

[0102] In this embodiment, before performing crossover mutation on the input population, in order to select the superior individuals from the input population and thus ensure that the mutation range is within the preferred range and accelerate the iteration speed, it is necessary to select the superior individuals through fast non-dominated sorting. Therefore, this embodiment also involves the following design:

[0103] The step of performing a fast non-dominated sort on the input population along the time and resource consumption dimensions, and selecting high-quality individuals from the fast non-dominated sort as the first optimized population, specifically includes:

[0104] Obtain the dominance relationship between each individual in the input population and other individuals in terms of duration and resource consumption. Divide all individuals in the input population into levels according to the number of other individuals they dominate, and calculate the crowding degree for each level. Select the high-quality individuals based on the level division and crowding degree calculation.

[0105] In an optional embodiment, the dominance relationship between each individual and other individuals in terms of duration and resource consumption is as follows: when one individual is better than another individual in both minimizing duration and minimizing resource consumption, it means that one individual dominates the other individual; if one individual is better than the other in only one of the two objectives of minimizing duration and minimizing resource consumption, there is no dominance relationship between the two individuals; all individuals are sorted according to the number of other individuals they can dominate, and the more individuals they can dominate, the better the individual is; individuals who can dominate the same number of individuals are divided into the same Pareto level.

[0106] Crowding is calculated for each pareto level. Crowding is the distance between an individual and its two neighboring individuals in the same pareto level. The closer the distance, the higher the crowding. Individuals with high crowding are considered superior in the same pareto level. By performing rapid non-dominated sorting and crowding calculation on the population, superior individuals are selected to enter the next generation.

[0107] It should be noted that in this embodiment, the mutation of the first optimized population is performed by generating new individuals through differential evolution operators. The mutation methods include crossover mutation and mixed mutation. Since the offspring population is obtained by crossover mutation of the optimized first optimized population, the probability of individuals in the offspring population being superior to previous individuals is higher. After multiple iterations, the population gradually converges towards the optimization goal. In this embodiment, the mutation method is as follows:

[0108] Since the probability matrix of the distribution estimation algorithm depicts the distribution of elite individuals, and the new population is generated by sampling the probability matrix, the population updates rapidly in the early stages of the algorithm, resulting in significant differences in the probability matrices between generations. However, in the later stages of the algorithm, the population update rate decreases, and the differences in the probability matrices between generations tend to stabilize. Therefore, based on this characteristic, the similarity coefficient of the probability matrices between two generations is used to compare the similarity of the two populations, thereby inferring the algorithm's running stage and selecting the appropriate mutation strategy. Assuming the algorithm is in the g-th iteration, the similarity coefficient is calculated as follows:

[0109]

[0110]

[0111] similar = (similar) JF +similar PA ) / 2;

[0112] As the algorithm converges in its later stages, the probability model gradually stabilizes, and the similarity value gradually decreases. Therefore, the following strategy is selected:

[0113]

[0114] Where τ represents the mutation strategy used, and rand(0,1) is a random number between 0 and 1.

[0115] The specific steps of the mutation strategy are as follows, and the mutation process of an example scheme is as follows: Figure 5 As shown, the probability of mutation is assumed to be 0.4.

[0116] First, the objective function of the population is calculated and a fast non-dominated sort is performed.

[0117] 1. Take one individual from the top 50% and one from the bottom 50% of the parent population, and represent their child task sequence vectors as X respectively. a X b Then, different individuals X are randomly selected from the population. c .

[0118] 2. Place X a X b Subtracting the elements in each dimension in digits, we get (X) a -X b ).

[0119] 3. Generate a random number rand in the interval [0,1], and select the value based on the scaling factor (X). a -X b )′, as shown in equation (3-25).

[0120]

[0121] Where X aj It is the subtask sequence vector X a The j-th dimensional gene locus.

[0122] 4. (X) a -X b )′ and X c The elements of each dimension are added in pairs. To prevent out-of-bounds errors, the remainder after dividing by the total number of subtasks n is incremented by 1 to obtain V′. i Mathematical expression: V′ ij =((X) aj -X bj )′+X ci )%n+1.

[0123] 5. V′ i By clearing the negative elements to zero, we can obtain the final sub-task vector V of the mutated individual. i ,Finish.

[0124] This embodiment further optimizes each individual in the second optimized population by performing variable neighborhood local search optimization in terms of duration and resource consumption, resulting in a third optimized population and improving the convergence efficiency of the population. It should be noted that, in order to further improve the effect of neighborhood local search optimization, this embodiment first extracts a set of superior individuals from the second optimized population according to preset conditions, and uses the superior individuals in this set to perform subsequent variable neighborhood local search optimization. The preset conditions are as follows:

[0125] The second optimized population is subjected to fast non-dominated sorting to obtain the set of non-dominated solutions in the second optimized population. Then, the other individuals in the second optimized population except the set of non-dominated solutions are sorted by single objective, that is, sorted by time dimension and resource consumption dimension respectively, and the top preset proportion of individuals are taken as the set of single objective optimal solutions. The set of non-dominated solutions and the set of single objective optimal solutions are used together as the set of excellent individuals.

[0126] The variable neighborhood local search optimization for the duration dimension is as follows:

[0127] The main task that occupies the most total sensor time and consumes the most total resources is selected as the first main task. The subtask with the latest completion time among all subtasks of the first main task is selected as the first subtask, and the first subtask is executed first.

[0128] By processing the subtasks that would otherwise take the longest and be executed the latest in advance, the overall task completion time can be reduced, thus avoiding the situation where the subtask alone slows down the overall time.

[0129] The variable neighborhood local search optimization for the resource consumption dimension is as follows:

[0130] The unmanned platform that consumes the most total resources is selected, and the subtask with the most resource consumption in the main task corresponding to the unmanned platform is taken as the second subtask. All sensors in the unmanned platform that can execute the second subtask are obtained, and the sensor with the least resource consumption for executing the second subtask is matched with the second subtask.

[0131] In addition to the variable neighborhood local search optimizations based on the duration and resource consumption dimensions mentioned above, this embodiment also includes a random variable neighborhood local search optimization, as follows:

[0132] Randomly select one or more unmanned platforms, randomly select a fourth preset proportion of sub-tasks from the main tasks corresponding to the unmanned platforms, reselect executable sensors for the fourth preset proportion of sub-tasks, and match the codes accordingly.

[0133] The fourth preset ratio is set by those skilled in the art based on the actual situation. The above-mentioned random variable neighborhood local search optimization is mainly to increase the randomness of the whole process, and the randomly selected individual may be the optimal solution.

[0134] In this embodiment, the relevant code for the variable neighborhood local search is as follows:

[0135]

[0136]

[0137] After each iteration, the third optimized population obtained at the end of the iteration needs to be transferred to an external archive set. Then, the external archive set is subjected to a fast non-dominated sort in terms of duration and resource consumption, thus obtaining the output individual in the external archive set at this time. This output individual is the relatively optimal individual before the end of this iteration. However, this individual may not necessarily meet the final optimization goal. The final optimization goal may require more iterations to be generated. Therefore, this embodiment needs to set the termination condition of the iteration according to the final optimization goal, as follows:

[0138] The termination conditions specifically include:

[0139] Set a preset number of iterations: Determine whether the current iteration round is greater than or equal to the preset number of iterations; when the current iteration round is greater than or equal to the preset number of iterations, the termination condition is met; when the current iteration round is less than the preset number of iterations, the termination condition is not met.

[0140] To further accelerate the convergence efficiency of genetic iteration, this embodiment updates the probability distribution model based on the latest external archive set after each iteration, so that the input population generated by the probability distribution model in the next iteration can get closer to the optimization target. Therefore, this embodiment also involves the following design:

[0141] The step of updating the probability distribution model based on the external archive set specifically includes:

[0142] The update formula for the corresponding probability between the main task code and the unmanned platform code is as follows:

[0143]

[0144] The update formula for the corresponding probability between subtask codes and sensors is as follows:

[0145]

[0146] Among them, JF i,j(g+1) represents the probability corresponding to the main task code i and the unmanned platform code j in the (g+1)th iteration; JF i,j (g) represents the probabilities corresponding to the main task code i and the unmanned platform code j in the g-th iteration; PA c,d (g+1) represents the probabilities corresponding to the subtask code c and the sensor code d in the (g+1)th iteration; PA c,d (g) represents the probability corresponding to subtask code c and sensor code d in the g-th iteration; α is the first preset learning rate, α∈(0,1); β is the second preset learning rate, β∈(0,1); Bestsize is the size of the external archive set.

[0147] For the external archive set, according to the indicator function of the Z-th elite individual in the fast non-dominated sorting, when the main task code i and the unmanned platform code j correspond in the g-th iteration, When the main task code i and the unmanned platform code j do not correspond in the g-th iteration,

[0148] For the external archive set, the indicative function of the Z-th elite individual in the fast non-dominated sorting is given when the subtask code c and the sensor code d correspond in the g-th iteration. When the subtask code c and the sensor code d do not correspond in the g-th iteration...

[0149] In this embodiment, the elite individuals are those selected from the top preset number or preset proportion of individuals after performing a fast non-dominated sort on the external archive set, which are then used by the probability distribution model for learning.

[0150] Since the probabilities in the probability distribution model are uniformly distributed in the initial iteration, optimization is performed after each iteration based on an external archive set. The optimal individual is selected from the external archive set using a fast non-dominated sorting algorithm. The corresponding probabilities of this individual are then used to optimize the probability distribution model using the formula described above, resulting in new probabilities. This increases the probability that individuals in the new population generated in the next iteration will be superior individuals. In this embodiment, both the first preset learning rate and the second preset learning rate are set by those skilled in the art based on actual conditions.

[0151] In this embodiment, after obtaining the output individual, the correspondence between the subtask and the sensor is displayed in encoded form as follows:

[0152] All subtask encoding sequences are arranged into a sequence vector according to their execution order. Then, based on the subtasks that each sensor in each unmanned platform needs to execute, these subtask codes are extracted and arranged and concatenated according to the order in the sequence vector. The concatenated sequence string is the scheduling string for that sensor; for example... Figure 6 As shown, specific examples are given below:

[0153] Figure 6 The sequence string at the top is a sequence string vector, where sensor F1 is responsible for executing subtasks 2, 3 and 9. Therefore, the above three subtasks are extracted from the sequence string vector and arranged and concatenated according to their order in the sequence string vector to obtain the scheduling string v1-0 of sensor F1. Figure 6 The scheduling string v2-0 for sensor F2 and the scheduling string v3-0 for sensor F3 are obtained in the above manner, and will not be elaborated further here.

[0154] Example 2:

[0155] like Figure 7 The diagram shown is a schematic of an unmanned swarm multi-sensor task allocation and scheduling optimization device according to an embodiment of the present invention. This unmanned swarm multi-sensor task allocation and scheduling optimization device includes one or more processors 41 and a memory 42. Figure 7 Take a processor 41 as an example.

[0156] Processor 41 and memory 42 can be connected via a bus or other means. Figure 7 Taking the example of a connection between China and Israel via a bus.

[0157] The memory 42, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs and non-volatile computer-executable programs, such as the unmanned swarm multi-sensor task allocation and scheduling optimization method in Embodiment 1. The processor 41 executes the unmanned swarm multi-sensor task allocation and scheduling optimization method by running the non-volatile software programs and instructions stored in the memory 42.

[0158] Memory 42 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, memory 42 may optionally include memory remotely located relative to processor 41, which can be connected to processor 41 via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0159] The program instructions / modules are stored in the memory 42. When executed by one or more processors 41, they perform the unmanned swarm multi-sensor task allocation and scheduling optimization method described in Embodiment 1 above. For example, they perform the above-described... Figure 1 , Figure 3 and Figure 4 The steps shown.

[0160] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for optimizing the allocation and scheduling of multi-sensor tasks in an unmanned swarm, characterized in that, include: The process involves acquiring unmanned cluster information and task information, encoding the unmanned cluster information and task information to obtain encoded information, generating an initial population through a probability distribution model and the encoded information, generating a correspondence between the main task code and the unmanned platform code in each individual based on the probability distribution model, generating a correspondence between the sub-task code and the sensor code in each individual based on the probability distribution model, thereby generating a single individual, and using multiple randomly generated individuals as the initial population. The initial population is used as the input population and fed into the first iteration. Each iteration includes: performing fast non-dominated sorting on the input population in terms of duration and resource consumption, and selecting high-quality individuals from the fast non-dominated sort as the first optimized population; mutating the first optimized population to generate a offspring population, merging the offspring population with the first optimized population for optimization, and obtaining the second optimized population; performing variable neighborhood local search optimization on each individual in the second optimized population in terms of duration and resource consumption, and obtaining the third optimized population; transferring the third optimized population to an external archive set, performing fast non-dominated sorting on all individuals in the external archive set to obtain the output individual, and determining whether the termination condition is met; When the termination condition is not met, the probability distribution model is updated according to the external archive set. The updated probability distribution model generates a new input population and is input into the next iteration. The update formula for the corresponding probability between the main task code and the unmanned platform code is as follows: ; The update formula for the corresponding probability between subtask codes and sensors is as follows: ； in, Let i be the probability corresponding to the main task code i and the unmanned platform code j in the (g+1)th iteration. Let i be the probability corresponding to the main task code i and the unmanned platform code j in the g-th iteration; Let c be the probability corresponding to the subtask code c and the sensor code d in the (g+1)th iteration. Let c be the probability corresponding to the subtask code c and the sensor code d in the g-th iteration; The first preset learning rate, ; The second preset learning rate, Bestsize is the size of the external archive set. For the external archive set, according to the indicator function of the Z-th elite individual in the fast non-dominated sorting, when the main task code i and the unmanned platform code j correspond in the g-th iteration, When the main task code i and the unmanned platform code j do not correspond in the g-th iteration, ; For the external archive set, the indicative function of the Z-th elite individual in the fast non-dominated sorting is given when the subtask code c and the sensor code d correspond in the g-th iteration. When the subtask code c and the sensor code d do not correspond in the g-th iteration, ; When the termination condition is met, the output individual is decoded to obtain the optimal solution.

2. The unmanned swarm multi-sensor task allocation and scheduling optimization method according to claim 1, characterized in that, The process of acquiring unmanned cluster information and task information, and encoding the unmanned cluster information and task information, specifically includes: Based on the unmanned cluster information, obtain information about all unmanned platforms; based on the information about all unmanned platforms, obtain information about each sensor in each unmanned platform; based on the task information, obtain information about all main tasks; based on the information about all main tasks, obtain information about each subtask in each main task. The unmanned platform code is obtained based on the information of all unmanned platforms; the sensor code is obtained based on the information of all sensors; the main task code is obtained based on the information of all main tasks; and the subtask code is obtained based on the information of all subtasks.

3. The unmanned swarm multi-sensor task allocation and scheduling optimization method according to claim 2, characterized in that, The generation of the initial population through the probability distribution model and the encoded information specifically includes: The initial population includes: individuals of a first preset proportion generated randomly, individuals of a second preset proportion generated using a resource consumption strategy, and individuals of a third preset proportion generated using a duration strategy.

4. The unmanned swarm multi-sensor task allocation and scheduling optimization method according to claim 3, characterized in that, The resource consumption strategy specifically includes: When generating the correspondence between subtask codes and sensor codes, among all the sensors available for the subtask, the sensor code of the sensor with the least resource consumption is matched with the subtask code first, thereby obtaining the second preset ratio of individuals.

5. The unmanned swarm multi-sensor task allocation and scheduling optimization method according to claim 3, characterized in that, The duration strategy specifically includes: When generating the correspondence between subtask codes and sensor codes, among all the sensors available for the subtask, the sensor that has been occupied by other subtasks for the least amount of time is selected as the first type of sensor. When the number of the first type of sensor is at most one, the sensor code and subtask code of the first type of sensor are matched to obtain the third preset ratio of individuals; When the number of the first type of sensors is at least two, the sensor code and subtask code of the sensor with the shortest consumption time are matched first among the first type of sensors to obtain the third preset ratio of individuals.

6. The unmanned swarm multi-sensor task allocation and scheduling optimization method according to claim 3, characterized in that, The step of performing a fast non-dominated sort on the input population along the time and resource consumption dimensions, and selecting high-quality individuals from the fast non-dominated sort as the first optimized population, specifically includes: Obtain the dominance relationship between each individual in the input population and other individuals in terms of duration and resource consumption. Divide all individuals in the input population into levels according to the number of other individuals they dominate, and calculate the crowding degree for each level. Select the high-quality individuals based on the level division and crowding degree calculation.

7. The unmanned swarm multi-sensor task allocation and scheduling optimization method according to any one of claims 1-6, characterized in that, The process of performing variable neighborhood local search optimization on each individual in the second optimized population in terms of duration and resource consumption to obtain the third optimized population specifically includes: The main task that occupies the most total sensor time and consumes the most total resources is selected as the first main task. The subtask with the latest completion time among all subtasks of the first main task is selected as the first subtask. The first subtask is executed first. Select the unmanned platform that consumes the most total resources, and take the subtask with the most resource consumption in the main task corresponding to the unmanned platform as the second subtask. Obtain all sensors in the unmanned platform that can execute the second subtask, and associate the sensor with the least resource consumption in executing the second subtask with the second subtask. Randomly select one or more unmanned platforms, randomly select a fourth preset proportion of sub-tasks from the main tasks corresponding to the unmanned platforms, reselect executable sensors for all the sub-tasks of the fourth preset proportion, and match the codes accordingly.

8. The unmanned swarm multi-sensor task allocation and scheduling optimization method according to any one of claims 1-6, characterized in that, The termination conditions specifically include: Set the preset number of iterations: Determine if the current iteration number is greater than or equal to the preset iteration number; The termination condition is met when the current iteration number is greater than or equal to the preset iteration number. The termination condition is not met when the current iteration number is less than the preset number of iterations.

9. A multi-sensor task allocation, scheduling, and optimization device for unmanned swarms, characterized in that, It includes at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the processor for performing the unmanned swarm multi-sensor task allocation and scheduling optimization method according to any one of claims 1-8.

Images