A sequence planning method for space large-scale target flyby based on parallel architecture

Abstract

The application discloses an optimal flyby sequence acquisition method based on a parallel architecture and belongs to the field of orbit design and optimization. The application realizes the method as follows: a parallel cluster search algorithm is constructed, and a Lambert problem is solved in parallel; a state of a spacecraft at a starting moment of a task time window is determined, a parallel cluster search is performed to obtain a sub-sequence of target access; then, the sub-sequence is saved to a database, a state and a moment at an end of each sub-sequence in the database are extracted, and the parallel cluster search is performed again from the state; the above process is repeatedly iterated to generate a sub-sequence database; the sub-sequences are spliced to generate a complete target access flyby sequence trajectory; and the total speed increment of the flyby sequence is further reduced by adjusting a transfer time between targets and optimizing a flyby speed. The application can significantly improve flyby sequence planning efficiency, effectively guarantee the optimality of sequence generation for a large-scale target problem, and acquire a high-precision and low-energy-consumption spacecraft flyby sequence.

Description

TECHNICAL FIELD

[0001] The application relates to a sequence planning method for space large-scale target flyby based on a parallel architecture and belongs to the field of space orbit design and optimization. BACKGROUND

[0002] The sequence planning for space target flyby refers to selecting a sequence of a series of space targets or planning the flyby order of all the targets, and optimizing the total speed increment of the whole flyby sequence under the constraint of flyby conditions. The targets of the task include natural celestial bodies such as planets and asteroids, and artificial celestial bodies such as probes and space debris, and the task involves small asteroid sequence access, space debris sequence capture, and kinetic energy interception on non-cooperative targets. For the sequence planning problem of a small number of targets, the existing technology can obtain the optimal flyby sequence through the exhaustion method. For the large-scale target sequence planning problem, as the number of targets increases, it is more difficult to obtain the optimal flyby sequence, the calculation complexity will have a dimension explosion, the greedy algorithm can effectively reduce the calculation burden, but it is easy to fall into a local optimal solution, and the beam search algorithm can improve the optimality of the problem result, but it will bring a large calculation burden. The existing technology lacks an algorithm for obtaining a better flyby sequence with high calculation efficiency. SUMMARY

[0003] For the sequence planning problem of space large-scale target flyby, the main purpose of the application is to provide an optimal flyby sequence acquisition method based on a parallel architecture. The large-scale Lambert problem is solved in parallel by constructing a parallel beam search algorithm, which effectively improves the solving efficiency of the large-scale Lambert problem in sequence planning. Given the number of flyby targets and the flyby time window required by the flyby task, the parallel beam search algorithm is used to quickly establish a flyby sub-sequence database, and the beam width is set in the beam search algorithm to splice the sub-sequences in the database under a wide enough beam width, which can effectively avoid falling into a local optimal solution in the flyby sub-sequence splicing process and improve the optimality of the whole sequence acquisition. The transfer time between targets in the complete sequence after splicing is adjusted and optimized through a double-layer optimization strategy, the total speed increment is reduced, and the optimality of the whole sequence is further improved. The flyby speed of each target in the sequence after the adjustment time is optimized by the nonlinear programming algorithm SQP, the total speed increment of the whole flyby process is further reduced, and the optimality of the whole sequence is further improved. Therefore, the application realizes the sequence planning for space large-scale target flyby based on a parallel architecture, and has the advantages of high planning efficiency and high optimality of flyby sequence planning.

[0004] The application is achieved by the following technical solutions.

[0005] The application discloses a sequence planning method for space large-scale target flyby based on a parallel architecture.

[0006] The application discloses a sequence planning method for space large-scale target flyby based on a parallel architecture.

[0007] Step one: in view of the problems of low optimization efficiency and difficulty in guaranteeing the optimality of planning results of the space large-scale target flyby sequence planning, the Lambert problem is solved based on a parallel architecture, and different input quantities are synchronously calculated to obtain respective results; after the Lambert problem calculation results are obtained, the speed increments corresponding to different input quantities are calculated in parallel, and a parallel cluster search algorithm is constructed with the speed increment as an index, the algorithm supports smaller search steps to improve the optimality of the obtained subsequence and improve the solving efficiency of the space large-scale Lambert problem.

[0008] Based on the powerful parallel computing capability of a graphic processing unit (GPU), first, the solving of the Lambert problem is established in a kernel function of the parallel architecture, that is, the Lambert problem is solved on each thread of the GPU; each thread can solve the corresponding Lambert problem for different inputs and output the corresponding results, and the calculation process is synchronously performed. The input quantity of the Lambert problem in each thread includes transfer time t f, initial position R0, terminal desired position R a , the output quantity includes the required velocity of the initial position The required velocity of the terminal desired position For a single-step flyby problem from one target to another, the velocity corresponding to the starting position is V0; therefore, the velocity increment vector to be applied at the starting position is The size of the velocity increment is Let the velocity of the terminal target be V a The minimum relative velocity limit for the flyby mission is Δv l The velocity increment vector Δv f is calculated by the piecewise function (1):

[0009]

[0010] The size of the arrival velocity increment is Δv f =||Δv f ||; therefore, for a single-step flyby problem, the total velocity increment size Δv is:

[0011] Δv = Δv0 + Δv f (2)

[0012] Based on the above-established parallel Lambert problem, it is applied to the parallel cluster search algorithm generated by the sub-sequence database. The parallel cluster search algorithm obtains the starting velocity and the terminal velocity of the transfer trajectory by solving the parallel Lambert problem under different input conditions, calculates the total velocity increment under each input condition by formula (2), and sets the cluster search width to screen the corresponding sub-sequence.

[0013] The parallel cluster search algorithm needs to allocate the corresponding grid number G s and block number B s to the parallel solving framework according to the number of Lambert problems to be solved; when there are N Lambert problems to be solved, the relationship between B s and G s is obtained by the following formula:

[0014]

[0015] Unlike the greedy algorithm, the parallel cluster search algorithm can retain the top B w optimal nodes by setting the hyperparameter B w , effectively avoid falling into a local optimal solution too early, and also reduce the space and time occupied by the search, and improve the solving efficiency of large-scale target Lambert problems in space.

[0016] Step 2: Given the number of targets to be flown and the flown time window required for the flyover mission, the transfer time between targets is discretized. Starting from the beginning of the time window, a subsequence is generated based on the parallel bundle search algorithm. By extracting the terminal state and time of the subsequence, the algorithm continues to generate subsequences, iterating repeatedly until the starting time is greater than the end value of the flyover time window. Based on the parallel bundle search algorithm, a flyover subsequence database can be quickly established, the database can be generated offline, and it can guide the generation of multiple flyover sequences.

[0017] The elements of the subsequence include target number, fly-over time, fly-over speed, and the magnitude of the speed increment from the previous target to the current target.

[0018] For N t This problem involves sequence planning for N possible objectives. Given the start time of a mission, we obtain the corresponding position and velocity. The spacecraft departs from this time, and there are N objectives to choose from. t Once the selected target completes its flight, the next candidate target will be N. t -1. Therefore, for the requirement to fly over N... flyby For a single fly-through mission, when the transfer time between targets is a fixed value, the total number of candidate sequences P is:

[0019]

[0020] In the formula, N t Choose N from the possible objectives. flyby Sort the elements.

[0021] For N t For smaller, less-than-ideal flyby sequence planning problems, the optimal visit sequence can be obtained through exhaustive search. However, for N... t A significant problem is that the total number of sequences generated by the exhaustive method exceeds the solvability of a computer; furthermore, the transition time between each objective is also a variable to be optimized and cannot be obtained in advance. Therefore, step two will establish a subsequence database generation method, which can effectively improve the generation efficiency of the optimal sequence.

[0022] Define the time window for the sequence flyby as [T0, T... f ], where T0 is the start time of the flyby mission, T f The final moment of the flyby mission, i.e., when the spacecraft needs to reach T... f Complete the N operation within the time limit of -T0. flyby The target is to be flown over; the maximum transfer time between targets is defined as [value]. It should be less than T f -T0, the minimum transition time is denoted as The transfer time between targets For the transfer between targets, the discrete intervals need to be exhaustively searched to obtain the optimal solution, and the discrete intervals are denoted as The transfer time

[0023] The position and velocity vector corresponding to the starting time T0 of the spacecraft are defined as R0 and V0, which are calculated first by using the parallel cluster search algorithm in step one. The input quantities include the transfer time t f , the positions and velocities of N t targets at the time T0+t f , and the total number of Lambert problems T idx to be solved is:

[0024]

[0025] Each thread separately completes the solution of one Lambert problem, and the solving process is synchronized. After completing the Lambert solution, each thread needs to calculate the corresponding velocity increment using equation (2). After completing the velocity increment solution, the calculation results are transmitted back to the CPU, and then the velocity increments corresponding to the T idx Lambert problems are sorted from small to large. The first B w velocity increments and the corresponding targets and transfer times are extracted. After extraction, the parallel cluster search is performed from the above B w targets, and the total number of Lambert problems T idx to be solved at this time is:

[0026]

[0027] According to the above steps of each thread first solving the Lambert problem and then obtaining the velocity increment, after the solution is completed, the T idx velocity increments are sorted from small to large, and the first B w subsequences are extracted. This is iterated until B w subsequences of length L are completed, which are saved in the database. In order to reduce the computational load of subsequent subsequence generation, after completing the subsequence search, the velocity increments of each subsequence are summed, and the corresponding maximum velocity increment limit is set to eliminate subsequences that exceed the limit.

[0028] Since the total length of the sequence N flyby is greater than the length of the subsequence L, generating only B w subsequences of length L cannot meet the task indicators. The target at the end of each subsequence, the corresponding arrival velocity, and the time are extracted from the existing database. After extraction, the parallel cluster search is performed until a subsequence of length L is generated again. Therefore, the above process is iterated until the departure time is greater than Tf , the rapid establishment of the sub-sequence database is completed. Based on the parallel beam search algorithm, the flyby sub-sequence database can be quickly established, the database is generated offline, and multiple flyby sequences can be generated.

[0029] Step three: by setting the beam width in the beam search algorithm, the sub-sequences in the database constructed in step two are spliced under a wide enough beam width, effectively avoiding falling into a local optimal solution in the flyby sequence splicing process, improving the optimality of the entire sequence acquisition, and after splicing is completed, the sequence S1 with the smallest total speed increment is selected.

[0030] The number of targets required to flyby in the flyby mission is N flyby , and the length of each sub-sequence is L; therefore, to achieve flyby of N flyby targets, N I sub-sequences need to be spliced, where N I is:

[0031]

[0032] The beam search width for splicing sub-sequences is defined as B w2 , first search from the starting time, take out all sub-sequences corresponding to the starting time from the database, calculate the total speed increment of each sub-sequence, sort from small to large according to the total speed increment, and select the first B w2 sub-sequences; extract the end targets of the B w2 sub-sequences and the corresponding time, extract the target sub-sequences corresponding to this time from the database, splice this sub-sequence with the former, obtain a sub-sequence with a length of 2L-1, calculate the total speed increment of all obtained sub-sequences, then sort them from small to large, and select the first B w2 sub-sequences; according to the above method, continuously iterate to complete the splicing of N I sub-sequences, and select the complete sequence S1 with the smallest speed increment. This step sets the beam width in the beam search algorithm, splices the sub-sequences under a wide enough beam width, effectively avoids falling into a local optimal solution in the sequence splicing process, and improves the optimality of the entire sequence acquisition.

[0033] Step four: by constructing a double-layer optimization strategy, adjust and optimize the transfer time between each target and target in the complete sequence S1 obtained in step three to reduce the total speed increment of the complete sequence, further improve the optimality of the entire sequence, and obtain a flyby sequence S2 with a lower total speed increment.

[0034] The transfer time between each target in the complete sequence S1 obtained in step three is adjusted by a double-layer optimization strategy, and the method is as follows: inner-layer optimization: given any two adjacent targets in the sequence, the flyby time of the former target in the sequence is fixed, the total speed increment required by the entire sequence is recalculated by changing the flyby time of the latter target; it should be noted that changing the flyby time of the latter target will affect the transfer time of the two targets; outer-layer optimization: starting from the initial time of the spacecraft, the inner-layer optimization between two targets is sequentially performed until the last target in the sequence; when a round of outer-layer optimization of the entire sequence is completed, the total speed increment required by the current sequence is recorded. By continuously iterating the above-mentioned outer-layer optimization and inner-layer optimization strategies, when the total speed increment values of adjacent two times of outer-layer optimization differ by 0, this optimization step is ended, and a flyby sequence S2 with lower total speed increment is obtained.

[0035] Step five: taking the flyby speed of each target as the optimization variable, the total speed increment of the sequence as the optimization target, and the relative speed condition of flyby as the constraint condition, an optimization problem model is established, and the flyby speed of each target in the sequence S2 obtained in step four is optimized by the nonlinear programming algorithm SQP, so as to further reduce the total speed increment of the entire flyby process, improve the optimality of the entire sequence, and obtain a flyby sequence S3 with smaller speed increment.

[0036] In the process of establishing the sub-sequence database, the calculation of the speed increment is to compare whether the relative speed between the arrival speed of Lambert problem and the current speed of the target satisfies the flyby constraint, such as formula (1); if the arrival speed of Lambert problem and the current speed of the target satisfy the flyby condition, no speed increment is applied; if the arrival speed of Lambert problem and the current speed of the target do not satisfy the flyby condition, the speed increment is applied to adjust the flyby speed. The end speed increment is calculated by formula (1), and the establishment of the sub-sequence database has high optimization efficiency, but the method of formula (1) only considers the optimality of the current flyby between targets, lacks consideration of the influence on the subsequent flyby sequence, and makes the total speed increment of the sequence too large.

[0037] In this step, the flyby speed V i of each target is taken as the optimization variable, the total speed increment of the sequence is taken as the optimization target, and the relative speed condition of flyby is taken as the constraint condition, an optimization problem model is established, and the flyby speed of each asteroid in the sequence S2 is optimized by the SQP optimization algorithm to obtain a flyby sequence S3 with smaller total speed increment. The objective function is:

[0038]

[0039] The constraint condition is:

[0040]

[0041] wherein, V i is the velocity of the i-th target i V i is the flyby velocity of the spacecraft to the i-th target, V i-1 is the arrival velocity of the i-1-th target to the i-th target through Lambert transfer, V i is the departure velocity of the i-th target to the i+1-th target through Lambert transfer, where i=1,2,...,N flyby .

[0042] Solving formula (8) and (9) by a nonlinear programming algorithm SQP can optimize the flyby velocity of each target in the sequence S2 obtained in step four, further reduce the total velocity increment of the entire flyby process, improve the optimality of the entire sequence, and obtain a flyby sequence S3 with smaller velocity increment, that is, a sequence planning for space large-scale target flyby based on a parallel architecture, and the flyby sequence with smaller velocity is obtained through optimization.

[0043] It also includes step six: according to the flyby sequence S3 with smaller velocity increment obtained in step five, the spacecraft is subjected to a corresponding velocity increment at a corresponding time, and under the velocity constraints of a given flyby window and flyby time, a high-precision low-energy consumption spacecraft flyby mission is realized.

[0044] Advantages:

[0045] 1. The sequence planning method for space large-scale target flyby based on a parallel architecture can effectively improve the solving efficiency of large-scale Lambert problems by constructing a parallel bundle search algorithm, quickly establish a flyby sub-sequence database, and splice the sub-sequences in the database under a wide enough bundle width in the bundle search algorithm, so that the flyby sequence search process can effectively avoid falling into a local optimal solution in advance, and the optimality of the entire sequence obtained is improved.

[0046] 2. The sequence planning method for space large-scale target flyby based on a parallel architecture can adjust the transfer time between targets of the complete sequence spliced through a double-layer optimization strategy. First, the inner-layer optimization is performed on two adjacent targets in the sequence to adjust the transfer time between the targets to reduce the velocity increment, and then the outer-layer optimization is continuously performed on the inner-layer optimization, and the iteration is performed until the total velocity increment values of two outer-layer optimizations are different. The complete sequence with smaller velocity increment is obtained. The double-layer optimization strategy can reduce the total velocity increment by optimizing the transfer time between targets, and further improve the optimality of the entire sequence.

[0047] 3. The sequence planning method for space large-scale target flyby based on parallel architecture, which takes the flyby speed of each target as the optimization variable, takes the total speed increment of the sequence as the optimization target, takes the relative speed condition of flyby as the constraint condition, establishes the optimization problem model, and adjusts the flyby speed of each target in the sequence after the time is optimized by the nonlinear programming algorithm SQP, so as to further reduce the total speed increment of the whole flyby process and improve the optimality of the whole sequence.

[0048] 4. The sequence planning method for space large-scale target flyby based on parallel architecture, which gives the required flyby target quantity and flyby time window, realizes the sequence planning of space large-scale target flyby based on parallel architecture on the basis of the above beneficial effects 1, 2 and 3, applies the corresponding speed increment to the spacecraft at the corresponding moment according to the optimized flyby sequence, and realizes the high-precision and low-energy-consumption execution of the predetermined spacecraft flyby task under the speed constraint of the given flyby window and flyby moment. The application has the advantages of high planning efficiency and high optimality of flyby sequence planning, BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 Fig. 1 is the distribution of flyby candidate targets in the embodiment.

[0050] Figure 2 Fig. 1 is the distribution of flyby candidate targets in the embodiment.

[0051] Figure 3 Fig. 1 is the distribution of flyby candidate targets in the embodiment.

[0052] Figure 4 Fig. 1 is the distribution of flyby candidate targets in the embodiment.

[0053] Figure 5 Fig. 1 is the distribution of flyby candidate targets in the embodiment.

[0054] Figure 6 Fig. 1 is the distribution of flyby candidate targets in the embodiment. DETAILED DESCRIPTION

[0055] In order to better illustrate the purpose and advantages of the application, the content of the application will be further described in combination with the drawings and examples.

[0056] Embodiment 1:

[0057] The embodiment discloses the optimal speed increment flyby sequence acquisition for asteroid flyby task, generates a flyby sequence with optimal total speed increment of the sequence for the large-scale candidate target problem provided by the task, and meets the relative speed constraint of flyby and the flyby time window constraint.

[0058] The candidate targets of this embodiment are selected from the top 10,000 asteroids in mass provided by the 11th International Space Trajectory Design Competition, and the distribution of the 10,000 candidate asteroids is as shown in Figure 1 The task objective is to select 33 asteroids for flyby from the 10,000 asteroids in the given time window with the minimum velocity increment, the flyby time window is 16 years, the task start time is 96030MJD, the relative velocity constraint of flyby is about 2km / s, and the task objective is to minimize the total velocity increment of the sequence.

[0059] As shown in Figure 2 , the embodiment discloses a sequence planning method for large-scale asteroid flyby, and the specific steps are as follows:

[0060] Step 1: Parallelize the Lambert problem solution, use CUDA and GPU parallel computing platform, and establish the solution of the Lambert problem on each thread, that is, each thread can solve the corresponding Lambert problem for different inputs, and then output the corresponding results, and the calculation process is synchronized. For the Lambert problem in each thread, the input quantity includes transfer time t f , initial position R0, and terminal desired position R a , and the output quantity includes the velocity corresponding to the initial position and the velocity corresponding to the terminal desired position As shown in Figure 3 , the efficiency of the established parallel Lambert problem solution is compared with that of the serial algorithm, and the calculation of the two algorithms is performed on a personal computer with the following configuration: Intel Core i9-10900X 3.70GHz CPU, 32Gb RAM, NVIDIA GeForce RTX 2080Ti GPU. As can be seen from Figure 3 , the parallel efficiency is much higher than the serial efficiency for solving large-scale Lambert problems; for example, for solving 50×10 6 Lambert problems, the parallel Lambert algorithm takes about 1.5s, while directly solving the Lambert problem on the CPU takes about 250s.

[0061] After solving the Lambert problem, the total velocity increment for transferring between targets needs to be calculated, which includes the velocity increment to be applied at the initial position and the velocity increment at the flyby moment. The velocity increment to be applied at the initial position is , and the size is The two cases of arrival velocity increment are as shown in Figure 4 , by assigning Δv l =2 to equation (1), equation (1) is specified as equation (10), and the arrival velocity increment vector can be calculated by the following segmented function:

[0062]

[0063] After completing the above speed increment calculation, the calculation results in the parallel computing framework are transferred back to the CPU, and the hyperparameter B is set. w =2000, only the first 2000 better nodes are retained to reduce the space and time occupied by the search and improve the solution efficiency.

[0064] Step 2: Based on the above parallel cluster search algorithm, establish an asteroid flyby subsequence database, which includes subsequences from Earth to asteroids and subsequences of asteroid-to-asteroid transfers.

[0065] First, starting from the spacecraft's position and velocity R0, V0 at the initial time T0, a subsequence from Earth to the asteroid is established, with a subsequence length of 2. In this embodiment, the spacecraft departs from Earth at a departure time of T0 = 96030 MJD. The spacecraft's position and velocity at departure time can be obtained using Kepler's formula. The parallel cluster search algorithm from step 1 is then used for calculation, with inputs including the transition time t. f The maximum transfer time The minimum value of the day Day, discrete interval The day also includes T0+t f Given the positions and velocities of 10,000 targets at time 1, the total number of Lambert problems to be solved is:

[0066]

[0067] Therefore, after completing the Lambert calculation, calculate the velocity increment for each variable, and given a velocity increment amplitude of 3.0 km / s, extract the first B. w =2000 subsequences, and the subsequences that satisfy this constraint will be stored in the database. After completing the Earth-to-asteroid subsequence calculation, the next step is to calculate the asteroid-to-asteroid subsequence; in this embodiment, the asteroid-to-asteroid sequence length is set to 3, the subsequence velocity increment constraint amplitude is 1.1 km / s, and the maximum transfer time is... The minimum value of the day Day, discrete interval sky.

[0068] Extract the final target and corresponding flyby velocity from the Earth-to-asteroid subsequence in the database. Starting from this target and its corresponding flyby velocity, use a two-stage parallel cluster search algorithm to generate a subsequence of length 3. The subsequence contains the following elements: target number, flyby time, flyby velocity, and velocity increment from the previous target to the current target. The subsequence is stored as shown in the table below:

[0069] Table 1: Sub-sequence length 3 saving mode

[0070]

[0071] After the sub-sequence search is completed, the total velocity increment of the sub-sequence is calculated, and if it is less than the constraint amplitude 1.1 km / s, it is saved to the database. Through this step, iteration is repeated, and finally a sub-sequence database is built.

[0072] Step three: The sub-sequences in the sub-sequence database are spliced, and the flyby sequence with the optimal velocity increment is selected by using the bundle search algorithm.

[0073] The number of targets required for the flyby mission is N flyby = 33, wherein the sequence length from the Earth to the asteroid is 2, and the sequence length between the asteroids is 3; the initial target and velocity of the sub-sequence are the terminal target and velocity corresponding to the existing sub-sequence, therefore, after one sequence splicing, the sequence length is increased by 2; to achieve flyby of 33 targets, 17 sub-sequences need to be spliced. The bundle search width for sub-sequence splicing is set to B w2 = 30000, first search from the starting time, take out all the sub-sequences corresponding to the starting time from the database, calculate the total velocity increment of each sub-sequence, sort according to the total velocity increment from small to large, and select the first B w2 sub-sequences; according to the size of the total velocity increment after sequence splicing, screening and constant iteration are performed to complete the splicing of the sub-sequences, and finally a complete sequence S1 with the smallest velocity increment is selected, and the three-dimensional trajectory corresponding to S1 is shown in Figure 5 .

[0074] Step four: adjust the transfer time between each target and target in the complete sequence S1 obtained in step three, wherein the flyby time of the 33 asteroids in S1 and the velocity increment between the asteroids are shown in Table 2. A double-layer optimization strategy for sequence target transfer time adjustment is constructed, the transfer time between targets is optimized to reduce the total velocity increment, and a new flyby sequence S2 is obtained, wherein the flyby time of S2 is the same as that of S3 in Table 2.

[0075] Table 2: Sequence S1 and S3 record table

[0076]

[0077] Step five: the flyby speed of each asteroid in the sequence S2 is optimized by using the SQP optimization algorithm, and a new flyby sequence S3 is obtained. The variable definitions of the flyby speed, Lambert transfer speed, and target speed of each target are as follows Figure 6The optimization problem constructed according to the formula (8) and the formula (9) is optimized by using a nonlinear programming algorithm, and the total velocity increment is further reduced; the velocity increment between the targets after optimization is shown in Table 2 as the flyby sequence S3. This step does not change the flyby time, and the total velocity increment of the sequence is reduced by optimizing the flyby velocity.

[0078] In the above sequence planning method for large-scale target flyby, for the case of large-scale optional targets, the establishment of the sub-sequence database can effectively improve the optimality of the sequence obtained, and the sub-sequence database can be quickly generated relying on the parallel beam search algorithm; the sub-sequences in the database are spliced by the beam search algorithm to obtain the complete flyby sequence trajectory; finally, the flyby time is adjusted and the flyby velocity is optimized to further reduce the total velocity increment.

[0079] The above specific description further details the purpose, technical solutions and beneficial effects of the application. It should be understood that the above description is only a specific embodiment of the application and is not used to limit the protection scope of the application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the application should be included in the protection scope of the application.

Claims

1. A sequence planning method for spatial large-scale target flyby based on parallel architecture, characterized in that: Comprising the following steps, Step one: In order to solve the problem of low optimization efficiency and difficulty in guaranteeing the optimality of the planning result of space large-scale target flyby sequence planning, the Lambert problem is solved based on parallel architecture, and the corresponding results are calculated synchronously for different input quantities; After obtaining the Lambert problem calculation result, the speed increment corresponding to different input quantities is calculated in parallel, and a parallel cluster search algorithm is constructed with the speed increment as the index. This algorithm supports smaller search steps to improve the optimality of the sub-sequence acquisition and improve the solving efficiency of the space large-scale target Lambert problem; Step two: Given the number of flyby targets and the flyby time window required by the flyby task, the transfer time between targets is discretized. Starting from the beginning of the time window, a sub-sequence is generated based on the parallel cluster search algorithm. By extracting the sub-sequence terminal state and time, the algorithm is used to continue generating sub-sequences, and the iteration is repeated until the departure time is greater than the end value of the flyby time window. The parallel cluster search algorithm can quickly establish a flyby sub-sequence database, generate the database offline, and guide the generation of multiple flyby sequences; Step three: By setting the cluster width in the cluster search algorithm, the sub-sequences in the database constructed in step two are spliced under a wide enough cluster width, effectively avoiding falling into a local optimal solution in the flyby sequence splicing process, improving the optimality of the entire sequence, and selecting the sequence S1 with the smallest total speed increment after splicing is completed; Step four: By constructing a double-layer optimization strategy, the transfer time between each target in the complete sequence S1 obtained in step three is adjusted and optimized to reduce the total speed increment of the complete sequence, further improve the optimality of the entire sequence, and obtain a flyby sequence S2 with a lower total speed increment; Step five: Taking the flyby speed of each target as the optimization variable, the total speed increment of the sequence as the optimization target, and the relative speed condition of the flyby as the constraint condition, an optimization problem model is established. By using the nonlinear programming algorithm SQP, the flyby speed of each target in the sequence S2 obtained in step four is optimized to further reduce the total speed increment of the entire flyby process, improve the optimality of the entire sequence, and obtain a flyby sequence S3 with a smaller speed increment.

2. The method of claim 1, wherein: Step six: According to the optimized flyby sequence S3 with a smaller speed increment obtained in step five, the corresponding speed increment is applied to the spacecraft at the corresponding time, and the spacecraft flyby task with high precision and low energy consumption is realized under the given flyby window and flyby time speed constraint.

3. The sequence planning method for space large-scale target flyby based on parallel architecture according to claim 1 or 2, characterized in that: The implementation method of step one is Based on the powerful parallel computing ability of GPU, the solution of Lambert problem is firstly established in the kernel function of parallel architecture, that is, the Lambert problem is solved on each thread of GPU; each thread can solve the corresponding Lambert problem for different input and output the corresponding result, and the calculation process is synchronous; for the Lambert problem in each thread, the input quantity includes the transfer time t f , the initial position R0, the terminal desired position R a , and the output quantity includes the required velocity at the initial position and the required velocity at the terminal desired position For the single-step flyby problem from one target to another, the velocity corresponding to the starting position is V0; therefore, the velocity increment vector to be applied at the starting position is The size of the velocity increment is Let the velocity of the terminal target be V a , and the minimum relative velocity limit of the flyby task be Δv l , then the velocity increment vector Δv f is calculated by the piecewise function (1): The size of the velocity increment Δv f = ||Δv f ||; therefore, for the single-step flyby problem, the total velocity increment size Δv is: Δv = Δv0+ Δv f (2) Based on the parallel Lambert problem established above, it is applied to the parallel cluster search algorithm for generating the sub-sequence database. The parallel cluster search algorithm solves the parallel Lambert problem under different input conditions to obtain the initial speed and terminal speed of the transfer trajectory. The total speed increment under each input condition is calculated by formula (2), and the cluster search width is set to select the corresponding sub-sequence; The parallel cluster search algorithm needs to assign the parallel solving framework with corresponding grid number G according to the number of Lambert problems to be solved s and block number B s ; when there are N Lambert problems to be solved, the relationship between B s and G s is obtained by the following formula: Unlike the greedy algorithm, the parallel beam search algorithm can effectively avoid falling into local optimal solution by setting the hyperparameter B w to retain the top B w optimal nodes, while also reducing the space and time occupied by the search and improving the efficiency of solving large-scale target Lambert problems.

4. The method of claim 3, wherein: The implementation method of step two is The elements of the sub-sequence include target number, flyby time, flyby speed, and speed increment from the previous target to the current target; For the sequence planning problem containing N t optional targets, given the starting time of the task, the corresponding position and velocity, the spacecraft starts from this time, there are N t optional targets to be selected, after the selected target is completed, the next optional target will be N t -1; therefore, for the flyby task requiring N flyby flyby, when the transfer time between targets is a fixed value, the total number of alternatives P is: In the formula, point to N t select N flyby elements from the N For N t small-scale target flyby sequence planning problems, the optimal visiting sequence can be obtained by the exhaustive method; but for N t large problems, the total number of sequences by the exhaustive method is beyond the scope of computer solvable; and the transfer time between each target is also a variable to be optimized, which cannot be obtained in advance; Therefore, the current step two will establish a sub-sequence database generation method, which can effectively improve the generation efficiency of the optimal sequence; Define the time window for sequence fly-through as [T0, T f ], where T0 is the start time of the flyby mission, T f The final moment of the flyby mission, i.e., when the spacecraft needs to reach T... f Complete the N operation within the time limit of -T0. flyby The target is to be flown over; the maximum transfer time between targets is defined as [value]. It should be less than T f -T0, the minimum transition time is denoted as The transfer time between targets For the transfer between targets, an exhaustive traversal is required given a discrete interval to obtain the optimal solution. Let the discrete interval be denoted as . Then the transfer time The position and velocity vector corresponding to the initial time T0 of the spacecraft are defined as R0, V0. First, the parallel bundle search algorithm in step one is used for calculation. The input quantities include the transfer time t f , the position and velocity of N f targets at the time T0+t t , and the total number T idx of Lambert problems to be solved is: Each thread solves one of the Lamberts individually, and the solving process is synchronized. After the Lamberts are solved, each thread also needs to calculate the corresponding velocity increment using equation (2). After the velocity increment is solved, the results are transmitted back to the CPU, and then T idx The velocity increments corresponding to the group of Lamberts are sorted from small to large, and the first B w velocity increments and the corresponding target and transfer time are extracted. After the extraction, from the above B w The target is to carry out parallel bundle search, and at this time the total number of Lambert problems to be solved T idx is: According to the step of solving the Lambert problem and obtaining the velocity increment of each thread, after the solving is completed, T idx The group velocity increments are sorted from small to large, and the first B w subsequences are extracted, and the iteration is continuously performed until the B w length L subsequences are completed, and the subsequences are saved in a database; in order to reduce the calculation amount of subsequent subsequence generation, after the subsequence search is completed, the velocity increments of each subsequence are summed, a corresponding maximum velocity increment limit is set, and subsequences exceeding the limit are removed. Since the total length of the sequence N flyby is greater than the length of the subsequence L, only B w subsequences of length L cannot meet the task indicators; extract the target, the corresponding arrival speed, and the time at the end of each subsequence from the existing database; after the extraction is completed, perform a parallel cluster search until a subsequence of length L is generated again; Therefore, the above process is iterated constantly until the departure time is greater than T f , the rapid establishment of the sub-sequence database is completed; the flyby sub-sequence database can be rapidly established based on the parallel cluster search algorithm, the database is generated offline, and multiple flyby sequences can be generated.

5. The method of claim 4, wherein: Step three is implemented by, The number of targets N required for a flyby mission flyby Each sub-sequence has a length L; thus, to achieve a flyby of N flyby targets, N I sub-sequences need to be stitched together, where N I is: The bundle search width for sub-sequence splicing is defined as B w2 First, search from the starting time, take out all the sub-sequences corresponding to the starting time from the database, calculate the total speed increment of each sub-sequence, sort from small to large according to the total speed increment, and select the first B w2 sub-sequences; extract the end target and the corresponding time of the B w2 sub-sequences, extract the target sub-sequence corresponding to the time from the database, splice the sub-sequence with the former, obtain a sub-sequence with a length of 2L-1, calculate the total speed increment of all obtained sub-sequences, then sort from small to large, and select the first B w2 sub-sequences; according to the above method, continuously iterate to complete the splicing of N I sub-sequences, and select the complete sequence S1 with the smallest speed increment; this step sets the bundle width in the bundle search algorithm, so that the sub-sequences are spliced under a wide enough bundle width, effectively avoiding falling into a local optimal solution in the sequence splicing process, and improving the optimality of the entire sequence acquisition.

6. The method of claim 5, wherein: Step four is implemented by, By double-layer optimization strategy, the transfer time between each target and target in the complete sequence S1 obtained in step three is adjusted, and the implementation method is as follows: inner optimization: given any two adjacent targets in the sequence, fix the flyover time of the former target in the sequence, and re-calculate the total speed increment required by the entire sequence by changing the flyover time of the latter; It should be noted that changing the flyover time of the latter will affect the transfer time of the former and the latter; Outer optimization: starting from the starting time of the spacecraft, sequentially perform the inner optimization between two targets, until the last target in the sequence; When completing a round of outer optimization of the entire sequence, record the total speed increment required by the current sequence; By constantly iterating the above outer optimization and inner optimization strategy, when the total speed increment values of adjacent two outer optimizations are 0, this optimization step is ended, and a flyby sequence S2 with lower total speed increment is obtained.

7. The method of claim 6, wherein: Step five is implemented by, In the process of establishing the sub-sequence database, the calculation of the speed increment is to compare the relative speed between the arrival speed of Lambert problem and the current speed of the target whether it meets the flyby constraint, such as formula (1); If the arrival speed of Lambert problem and the current speed of the target meet the flyby condition, no speed increment is applied; If the arrival speed of Lambert problem and the current speed of the target do not meet the flyby condition, the speed increment is applied to adjust the flyby speed; The end speed increment is calculated by formula (1), and the establishment of the sub-sequence database has high optimization efficiency, but formula (1) only considers the optimality of the current target flyby, lacks consideration of the influence on the subsequent flyby sequence, and makes the total speed increment of the sequence too large; The step takes the flyby speed V of each target i As the variable to be optimized, the total speed increment of the sequence is taken as the optimization target, the relative speed condition of the flyby is taken as the constraint condition, an optimization problem model is established, the flyby speed of each asteroid in the sequence S2 is optimized through an SQP optimization algorithm, and a flyby sequence S3 with a smaller total speed increment is obtained; the objective function is: The constraint condition is: wherein, V is the velocity of the ith target, i V is the flyby velocity of the spacecraft to the ith target, V represents the arrival velocity of the Lambert transfer from the (i-1)th target to the ith target, V represents the departure velocity of the Lambert transfer from the ith target to the (i+1)th target, where i = 1, 2,..., N flyby ; By solving formula (8) and (9) through the nonlinear programming algorithm SQP, the flyby speed of each target in the sequence S2 obtained in step four can be optimized, the total speed increment of the entire flyby process is further reduced, the optimality of the entire sequence is improved, and a flyby sequence S3 with smaller speed increment is obtained, that is, a sequence planning for space large-scale target flyby based on parallel architecture is realized, and a flyby sequence with smaller speed is obtained through optimization.

Images