A multi-uuv replacement path planning method with adaptive dual operator and embedded refinement

By constructing a spatiotemporal reachability model and adaptive genetic-simulated annealing co-optimization in complex ocean current environments, the problem of unreachability when the parent platform and UUVs meet dynamically was solved, achieving efficient planning of multiple UUV replacement paths, reducing replacement time and improving solution accuracy.

CN123067449APending Publication Date: 2026-11-10HARBIN UNIV OF SCI & TECH
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202611113390.2
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-26
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing technologies struggle to achieve dynamic rendezvous between the mother platform and multiple unmanned underwater vehicles (UUVs) in complex ocean current environments, leading to unreachable path planning and inefficient replacement tasks. Furthermore, traditional algorithms are prone to getting trapped in local optima in large-scale tasks, making it difficult to balance solution accuracy and efficiency.

Method used

By constructing a spatiotemporal reachability model that integrates ocean current speed, mother platform navigation capability, and UUV motion status, an adaptive genetic search mechanism and simulated annealing collaborative optimization strategy are designed to dynamically adjust crossover and mutation probabilities. This is combined with targeted flip mutation based on high-time-consuming segment identification, and deep annealing and population reconstruction are triggered when the population stagnates, ensuring the reachability and efficiency of path planning.

Benefits of technology

It improves the accessibility and efficiency of multi-UUV replacement path planning, reduces replacement time, reduces the average time by 27.52% compared with traditional methods, and enhances global search capability and local optimization accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123067449A_ABST
    Figure CN123067449A_ABST
Patent Text Reader

Abstract

To solve the problem of kinematic deadlock and premature convergence of traditional path planning methods in ocean current, an adaptive dual-operator and embedded refinement multi-UUV replacement path planning method is proposed. Firstly, a deadlock filtering mechanism based on Zermelo equation is introduced to explicitly eliminate the unreachable reverse water segments and ensure the physical feasibility of the path. Secondly, an adaptive dual-operator mechanism driven by flow field characteristics is designed to protect the elite down-current path by adaptively adjusting the crossover probability and to accurately flip the high-time-consuming reverse water nodes by adaptively adjusting the mutation probability, effectively reducing the task time consumption. Finally, an embedded refinement mechanism is proposed to periodically trigger local refinement by simulated annealing and write the results back to the population, while performing gradient blood replacement reset in deep stagnation to maintain the global optimization activity. The simulation results show that the proposed method can significantly shorten the replacement path task time, improve the path planning ability in complex ocean current environment, and meet the multi-UUV dynamic replacement task requirements.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention belongs to the field of intelligent path planning technology for unmanned underwater vehicles (UUVs), and relates to dynamic replacement path planning for multiple UUVs in ocean current environments. In particular, it relates to an adaptive dual-operator and embedded refined closed-loop path planning method for multiple UUVs. Background Technology

[0002] Unmanned Underwater Vehicles (UUVs), as unmanned system platforms capable of autonomously performing underwater exploration and operational tasks, have been widely used in fields such as marine resource exploration, underwater target search, and marine environmental monitoring. Coordinated replacement of multiple UUVs by a parent platform is a crucial means of ensuring the continuous operational capability of UUV swarms. However, complex ocean currents in the marine environment can alter the parent platform's ground-based speed and the actual travel time for each segment, making it difficult for the parent platform and moving UUVs to accurately rendezvous, and even creating kinematically inaccessible paths, severely impacting the efficiency of replacement missions. Therefore, researching efficient dynamic replacement path planning methods for multiple UUVs is essential for improving the continuous operational capability of underwater unmanned systems. Current mainstream multi-UUV replacement path planning methods still have limitations in ocean current environments: traditional combinatorial optimization-based path planning methods usually simplify the segment cost to Euclidean distance or travel time at constant speed, making it difficult to accurately describe the relationship between ocean currents, the motion state of the mother platform, and the dynamic position of the UUVs, and easily generating access paths that are actually unreachable; path planning methods based on intelligent optimization algorithms have a certain global search capability, but they are prone to getting trapped in local optima, and it is difficult to balance solution accuracy and convergence efficiency when the number of UUVs increases.

[0003] Existing improvement schemes mostly focus on single-dimensional performance enhancements, failing to systematically address the comprehensive optimization problem among ocean current environment perception, dynamic rendezvous accessibility, and global-local collaborative optimization. Therefore, there is an urgent need to design a dynamic replacement path planning method for multiple unmanned underwater vehicles that features strong path accessibility, low replacement time, and high solution stability, in order to overcome the technical bottleneck of path planning for collaborative replacement tasks of the mother platform under complex ocean current conditions.

[0004] The paper "Analysis and Solution Method for TSP Problem Based on Path Contribution Evaluation of Two-Population Ant Colony Algorithm" has the following problems: This method mainly uses two independent ant colonies to perform fast convergence search and path contribution search respectively. The two colonies only cooperate through the comparison of the best path at each stage, and lack a dynamic scheduling mechanism based on the population fitness distribution and continuous stagnation state. At the same time, its search process mainly relies on pheromone updates and does not form a synergistic optimization between adaptive crossover, directional mutation and local refinement, making it difficult to further improve near-optimal paths in complex scenarios.

[0005] The paper "Research on Traveling Salesman Problem Based on Improved Exclusion Genetic Algorithm" has the following problems: The method mainly maintains population diversity through niche exclusion and path value preprocessing, but its crossover and mutation processes are not dynamically adjusted according to the differences in population fitness and the degree of continuous stagnation; at the same time, the path value preprocessing relies on fixed inter-node costs and lacks targeted optimization for dynamic high-time-consuming flight segments.

[0006] The paper "Solving the Generalized Traveling Salesman Problem Based on an Improved Artificial Bee Colony Algorithm" has the following problems: The method mainly adjusts the perturbation intensity adaptively through multiple neighborhood update rules and the number of stagnations. Its adaptive mechanism only acts on a single perturbation process and lacks the coordinated regulation of the crossover and mutation operators. At the same time, its ability to optimize the near-optimal access sequence in the later stage of the search is still insufficient.

[0007] The paper "An Improved Ant Colony Algorithm Based on MATLAB for Solving the Traveling Salesman Problem" has the following problems: The method improves the traditional ant colony algorithm through dynamic parameter adjustment, genetic population recombination and simulated annealing perturbation, but each optimization mechanism is mainly executed sequentially according to the preset stage, and lacks a dynamic switching mechanism based on the actual stagnation state of the population; simulated annealing is only used as a late perturbation method, and does not periodically refine the elite path locally and continuously write the optimization results back to the population, resulting in insufficient synergy between macro-global search and micro-local optimization.

[0008] Patent CN111612217A, "An Improved GA-SA Path Planning Method for Surface Unmanned Vessels for Offshore Wind Farm Inspection," has the following problems: The simulated annealing method mainly focuses on the current best individual to carry out routine neighborhood perturbation, does not distinguish between periodic elite refinement and deep stagnation rescue based on the degree of continuous stagnation, and also lacks directional flipping and population gradient reconstruction mechanisms for high-time-consuming segments against the current. Summary of the Invention

[0009] To address the problems existing in the prior art, this invention provides an adaptive dual-operator and embedded refined multi-UUV replacement path planning method. The aim is to solve the problems of kinematic unreachability of dynamic rendezvous points, easy getting trapped in local optima in the access order, and low efficiency in solving large-scale tasks in the replacement path planning of the mother platform under complex ocean current environments. First, a spatiotemporal reachability model integrating ocean current speed, the mother platform's navigation capability, and the dynamic motion state of the unmanned underwater vehicles (UUVs) is constructed to calculate the earliest feasible rendezvous time between the mother platform and each UUV in real time. Kinematic constraints are embedded into the path evaluation process to eliminate unreachable paths that do not meet the navigation speed limits. Second, an adaptive genetic search mechanism is designed to dynamically adjust the crossover and mutation probabilities based on the population fitness distribution and search status. Targeted flip mutation is implemented in conjunction with the identification results of time-consuming segments, enhancing population diversity while reducing additional time losses caused by unfavorable ocean current segments. Finally, a hybrid optimization strategy combining genetic search and simulated annealing is proposed. Periodic embedded simulated annealing is used to locally refine superior individuals, and deep annealing and population reconstruction are triggered when the population stagnates for a long period. This balances global exploration capability with local optimization accuracy, obtaining the optimal access path for the mother platform that satisfies dynamic rendezvous constraints and has the lowest total replacement time.

[0010] To achieve the above objectives, the present invention employs the following specific technical solutions:

[0011] An adaptive dual-operator and embedded refined multi-UUV replacement path planning method includes the following steps:

[0012] S1. Establish a system kinematic model of the parent platform and multiple UUVs, construct the dynamic cooperative replacement task as a moving target traveling salesman problem, and perform deadlock filtering. This includes the following sub-steps:

[0013] S11. Establish a global two-dimensional rectangular coordinate system Assume there are [illegible] sea areas. Let the set of UUVs to be replaced be denoted as . ,in, For UUV numbering, Indicates the first One UUV to be replaced; for any UUV Let the coordinates of the geometric center of its circumference be... ,in, and They represent the first The geometric center of the UUV cruising trajectory is at shaft and The coordinates along the axis, the travel radius is With constant waterline velocity Conduct a circular tour;

[0014] S12. Assume that a stable macroscopic ocean current field exists within the operating area, and its velocity vector is constant. ,in, and These represent the ocean current speeds at... shaft and Components in the axial direction; let This represents the actual velocity vector of the parent platform relative to the ground. The relative hydrostatic thrust velocity vector of the parent platform; The maximum powered speed of the mother platform in still water conditions, and the ground motion of the mother platform satisfies the Zermelo navigation equations:

[0015]

[0016] S13. In the process of dynamic collaborative replacement, let... This indicates the number of pursuit segments currently completed by the parent platform, at the specified time. vector from current position Set sail and pursue the target UUV Then the absolute spatial position coincidence constraint must be satisfied, that is:

[0017]

[0018] In the formula, Indicates the start time of the current pursuit segment. Indicates the parent platform at any time position vector, This indicates the candidate meeting point between the parent platform and the target UUV. Indicates the first At the moment of intersection, two UUVs The actual position vector;

[0019] Since the parent platform adopts a straight-line shortest approach strategy between the two nodes, the average velocity vector relative to the ground required for the parent platform to achieve this rendezvous can be derived:

[0020]

[0021] In the formula, This indicates that the candidate meeting time has reached the [number]th [time]. The average ground velocity vector required for each UUV;

[0022] Substituting equation (3) into equation (1), we can obtain the hydrostatic propulsion speed required by the mother platform during the approach process. :

[0023]

[0024] For the rendezvous to be physically feasible, the hydrostatic thrust required by the mother platform must strictly satisfy the kinematic boundary constraints:

[0025]

[0026] When ocean currents force the mother platform to approach in the opposite direction of the current, if at all feasible meeting moments... The required hydrostatic thrust exceeds the dynamic limit, i.e. If the condition is always true, then the alternative flight segment is determined to have a kinematic deadlock, and the access sequence is physically unreachable.

[0027] S14. Let the access sequence of the parent platform be... ,in Indicates that the parent platform is from UUV Depart and eventually return, for An arrangement describing the parent platform's... The access order of each UUV;

[0028] For sequences The first in One pursuit segment, assuming the mother platform is at a certain time... The replacement of the previous node has been completed; the current position is... Since the target UUV is in a continuous circular cruising state, the actual pursuit time of the parent platform cannot be obtained by calculating the static Euclidean distance. Instead, it is necessary to solve for the earliest feasible rendezvous time that satisfies the dynamic boundary conditions based on the Zermelo kinematic constraints.

[0029]

[0030] In the formula, Indicates the moment when candidates meet. Indicates the previous meeting point. Indicates the first The earliest feasible rendezvous time for each leg of the journey. Indicates the parent platform in the Position vector at the start of each flight segment Indicates the target UUV at time... The dynamic position vector;

[0031] If equation (6) has no solution over the entire time interval, an extreme value penalty is imposed on its fitness: , Fitness of the access sequence;

[0032] S15. Define the path planning problem as follows: under the conditions of satisfying Zermelo kinematic constraints and each UUV being visited exactly once, solve the problem to enable the parent platform to complete all... The mathematical model for the access sequence with the minimum total time for UUV replacement tasks is as follows:

[0033]

[0034]

[0035]

[0036] In the formula, Indicates access sequence The corresponding total replacement task time on the parent platform; Indicates the first Time taken for each pursuit segment; Indicates that the parent platform executes the first The relative still water propulsion speed vector required for each pursuit segment; for The set of all valid permutations of the above;

[0037] S16, the access sequence for each individual in the population. The earliest feasible rendezvous time is extracted segment by segment according to equation (6). The time spent on each flight segment Accumulate the total time consumed in the sequence and update the position of the parent platform segment by segment. ,in, This indicates that the parent platform has completed the first... The new position vector after replacing the target UUV; if there is no feasible rendezvous time in a certain flight segment throughout the entire time range, it is determined that a kinematic deadlock has been triggered, the recursion is immediately terminated and an extreme value penalty is applied, and it is excluded from the valid candidate set;

[0038] S2. Calculate the population fitness and dynamically monitor its evolutionary state, which includes the following sub-steps:

[0039] S21. Minimum task time for extracting valid individuals in the current generation. Compared with the average time of the population Define the evolutionary stagnation index:

[0040]

[0041] S22. Let the global historical best time consumption be... The previously recorded globally optimal time was ,like That is, the current generation discovers a shorter path to replace the parent platform, Update to the current optimal anti-flow field path. Continuous stagnation counter. The update formula is defined as follows:

[0042]

[0043] In the formula, For the set tolerance, if If the number of cases continues to accumulate and reaches a preset threshold, it is determined that the replacement path optimization of the parent platform has stalled, thereby triggering a switch between macro and micro mechanisms.

[0044] S3. Enter the adaptive genetics stage, dynamically adjusting the crossover and mutation probabilities of individuals, specifically including the following sub-steps:

[0045] S31. Select the generation with the shortest total time through a tournament to enter the parent pool; use a non-linear adaptive crossover probability based on individual fitness during the crossover phase. Let the time taken by the better parent among the two parent generations participating in the crossover be... The calculation rules are as follows:

[0046]

[0047] In the formula, and These are the upper and lower bounds of the crossover probability, respectively.

[0048] S32. In the mutation phase, an exponential adaptive mutation probability based on the population stagnation index is adopted. :

[0049]

[0050] In the formula, and These are the upper and lower bounds of the mutation probability, respectively. As a regulator of sensitivity to variation, It is an index of evolutionary stagnation.

[0051] S33. For each individual to be mutated, firstly, extract the time consumption data for each segment of its path, locate the upstream bottleneck segment where the mother platform approaches the UUV against the ocean current and has the highest time consumption, and then perform a flip operation on the local path segment centered on this segment, forcibly removing the high-time-consuming upstream segment from the mother platform's current access sequence. Compared to random flipping, this directional strategy can eliminate the risk of kinematic deadlock at the path level of ocean current interference in a more accurate way, and generate downstream candidate paths with lower resistance;

[0052] S4. Perform simulated annealing collaborative refining and deep stagnation processing, periodically refine the elite access sequence locally, and trigger deep simulated annealing and population reconstruction when the population is in deep stagnation. This includes the following sub-steps:

[0053] S41, When the continuous stop counter Reaching the set threshold At that time, a macro-micro switch is triggered to extract the current globally optimal access sequence. The deep simulated annealing algorithm is invoked as the initial solution. At each temperature, a neighborhood perturbation operation is randomly applied to the current parent platform access sequence to generate new candidate solutions. The neighborhood perturbation employs two alternating operations: first, 2-opt flipping, randomly selecting two positions in the sequence and flipping the UUV access subsequence between them; second, node swapping, randomly selecting two UUVs and swapping their order in the access sequence. The acceptance of candidate solutions follows the Metropolis criterion.

[0054]

[0055] In the formula, This indicates the current temperature during the simulated annealing process. This represents the difference in total time taken for the parent platform replacement path between the candidate solution and the current solution;

[0056] S42. After deep annealing, regardless of whether improvement is achieved, a gradient blood replacement operation is performed: the first position in the population is forcibly filled with the current globally optimal access sequence. ; 2nd to 3rd The next best elite individuals are filled in according to their fitness ranking. The number of elite individuals retained during the tiered blood transfusion process is determined by the elite retention ratio. With population size determined; the first To the Replace all bits with a randomly generated valid access sequence. This refers to the population size. The counter will remain continuously idle after the blood transfusion is complete. Zeroing out; in addition, during the normal evolution phase, every The generation replaces some individuals with lower fitness rankings in the population with randomly generated legal access sequences according to a random injection ratio, as a normal supplement to suppress homogenization;

[0057] S43. When deep stagnation is not triggered, and the current evolution generation is within the embedded refinement trigger cycle. When the value is an integer multiple of the target value, initiate embedded genetic-simulated annealing co-refining: Select The optimal parent platform replaces the UUV path to perform local refinement, and the dynamic initial temperature is calculated for each elite individual:

[0058]

[0059] In the formula, For temperature coefficient, Total time spent replacing the path for the current elite;

[0060] S44. Repeat the above steps until the maximum number of generations is reached. The iteration is terminated, and the optimal planning path and shortest total time for the parent platform to perform dynamic collaborative replacement tasks for multiple UUVs are output.

[0061] The present invention has the following beneficial effects:

[0062] (1) To address the issues of insufficient local optimization capabilities and the disconnect between macroscopic search and microscopic development in multi-UUV replacement tasks, this invention proposes an embedded refinement and macro-microscopic switching mechanism. By periodically calling the simulated annealing algorithm to locally refine the elite access sequence, and writing the refined superior results back to the genetic population, continuous feedback of local optimization results to the global evolutionary process is achieved. When the population enters a deep stagnation state, deep simulated annealing and gradient blood exchange operations are triggered to restore the population's global optimization vitality. In simulation experiments, the average replacement time of the method in this invention is reduced by 27.52% compared to the traditional genetic algorithm, providing reliable technical support for multi-UUV replacement path planning in ocean current environments.

[0063] (2) To address the problem that traditional genetic algorithms, which use fixed crossover and mutation probabilities, easily destroy superior path segments and struggle to escape local optima, this invention designs an adaptive dual-operator mechanism. Specifically, the crossover probability is dynamically adjusted based on the population fitness distribution to protect superior downstream path segments; the mutation probability is dynamically adjusted based on the degree of population evolution stagnation, and directional flip mutation is implemented in conjunction with the identification results of time-consuming flight segments. Through the synergistic effect of the adaptive crossover operator and the directional mutation operator, population diversity is enhanced while protecting superior individuals, improving the global search capability and multi-UUV replacement path optimization efficiency of this method. The average replacement time is 8.33% less than that of the traditional simulated annealing algorithm.

[0064] (3) To address the problem that traditional path planning methods in complex ocean current environments use static distance as the segment cost, which makes it difficult to accurately describe the rendezvous process between the mother platform and the dynamic unmanned underwater vehicle (UUV), this invention constructs a spatiotemporal reachability model that integrates ocean current speed, mother platform navigation capability, and UUV motion state. By calculating the earliest feasible rendezvous time between the mother platform and the target UUV and embedding kinematic constraints into the path evaluation process, unreachable segments that do not meet speed constraints can be effectively eliminated, improving the accuracy of segment time calculation and the actual feasibility of alternative paths. Attached Figure Description

[0065] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0066] Figure 1 This is the overall flowchart of the present invention;

[0067] Figure 2 This is a diagram showing the optimal replacement path for the simulated UUV in this invention;

[0068] Figure 3 Simulate the optimal replacement path graph for UUVs using a traditional genetic algorithm;

[0069] Figure 4 Simulate the optimal replacement path graph for UUVs using the traditional simulated annealing algorithm;

[0070] Figure 5 This is the cumulative replacement success curve of the present invention;

[0071] Figure 6 This is the cumulative replacement success curve for a traditional genetic algorithm.

[0072] Figure 7 This is the cumulative replacement success curve for the traditional simulated annealing algorithm. Detailed Implementation

[0073] The following is in conjunction with the appendix Figure 1 The method steps and technical details of the present invention will be further explained below;

[0074] This invention addresses the problems of traditional path planning methods in ocean current environments, such as lack of physical flow field perception, excessively long upstream journey times and susceptibility to kinematic deadlock, and premature convergence and failure in large-scale scenarios due to the separation of macro and micro search mechanisms in conventional hybrid heuristic algorithms. It provides an adaptive dual-operator and embedded refinement method for multi-UUV replacement path planning. By establishing a UUV circular cruising kinematic model and a parent platform Zermelo kinematic model, the dynamic cooperative replacement task of multiple UUVs by the parent platform is constructed as a moving target traveling salesman problem to solve the problem of lack of physical flow field perception. An adaptive dual-operator cooperative mechanism protects elite downstream paths and improves the high-time-consuming upstream journey times. An embedded genetic-simulated annealing cooperative refinement mechanism is integrated to achieve real-time feedback of micro-level local development results to macro-level evolution, ensuring path planning capability in large-scale ocean current fields. The specific steps are as follows:

[0075] S1. Establish a system kinematic model of the parent platform and multiple UUVs, construct the dynamic cooperative replacement task as a moving target traveling salesman problem, and perform deadlock filtering. This includes the following sub-steps:

[0076] S11. Establish a global two-dimensional rectangular coordinate system Assume there are [illegible] sea areas. Let the set of UUVs to be replaced be denoted as . ,in, For UUV numbering, Indicates the first One UUV to be replaced; for any UUV Let the coordinates of the geometric center of its circumference be... ,in, and They represent the first The geometric center of the UUV cruising trajectory is at shaft and The coordinates along the axis, the travel radius is With constant waterline velocity Conduct a circular tour;

[0077] S12. Assume that a stable macroscopic ocean current field exists within the operating area, and its velocity vector is constant. ,in, and These represent the ocean current speeds at... shaft and Components in the axial direction; let This represents the actual velocity vector of the parent platform relative to the ground. The relative hydrostatic thrust velocity vector of the parent platform; The maximum powered speed of the mother platform in still water conditions, and the ground motion of the mother platform satisfies the Zermelo navigation equations:

[0078]

[0079] S13. In the process of dynamic collaborative replacement, let... This indicates the number of pursuit segments currently completed by the parent platform, at the specified time. vector from current position Set sail and pursue the target UUV ;in, Indicates the start time of the current pursuit segment. Indicates the parent platform at any time Position vector; let For the candidate rendezvous moment between the parent platform and the target UUV, the absolute spatial position coincidence constraint must be satisfied, that is:

[0080]

[0081] In the formula, Indicates the start time of the current pursuit segment. Indicates the parent platform at any time position vector, This indicates the candidate meeting point between the parent platform and the target UUV. Indicates the first At the moment of intersection, two UUVs The actual position vector;

[0082] Since the parent platform adopts a straight-line shortest approach strategy between the two nodes, the average velocity vector relative to the ground required for the parent platform to achieve this rendezvous can be derived:

[0083]

[0084] In the formula, This indicates that the candidate meeting time has reached the [number]th [time]. The average ground velocity vector required for each UUV;

[0085] Substituting equation (3) into the Zermelo equation (1), we can obtain the hydrostatic propulsion speed required by the mother platform during the approach process. :

[0086]

[0087] For the rendezvous to be physically feasible, the hydrostatic thrust required by the mother platform must strictly satisfy the kinematic boundary constraints:

[0088]

[0089] When ocean currents force the mother platform to approach in the opposite direction of the current, if at all feasible meeting moments... The required hydrostatic thrust exceeds the dynamic limit, i.e. If the condition is always true, then the alternative flight segment is determined to have a kinematic deadlock, and the access sequence is physically unreachable.

[0090] S14. Let the access sequence of the parent platform be... ,in Indicates that the parent platform is from UUV Depart and eventually return, for An arrangement describing the parent platform's... The access order of each UUV;

[0091] For sequences The first in One pursuit segment, assuming the mother platform is at a certain time... The replacement of the previous node has been completed; the current position is... Since the target UUV is in a continuous circular cruising state, the actual pursuit time of the parent platform cannot be obtained by calculating the static Euclidean distance. Instead, it is necessary to solve for the earliest feasible rendezvous time that satisfies the dynamic boundary conditions based on the Zermelo kinematic constraints.

[0092]

[0093] In the formula, Indicates the moment when candidates meet. Indicates the previous meeting point. Indicates the first The earliest feasible rendezvous time for each leg of the journey. Indicates the parent platform in the Position vector at the start of each flight segment Indicates the target UUV at time... The dynamic position vector;

[0094] when When it exists, the time taken for this flight segment is recorded as... The location of the parent platform was then updated to This is used as the starting state for the next leg of the journey, and the process is repeated. If equation (6) has no solution over the entire time period, it means that the ocean current forces the required still water thrust for this leg to continuously exceed the power limit of the mother platform, resulting in a kinematic deadlock. The access sequence is physically unreachable, and an extreme value penalty is imposed on its fitness. , Fitness of the access sequence;

[0095] S15. Define the path planning problem as follows: under the conditions of satisfying Zermelo kinematic constraints and each UUV being visited exactly once, solve the problem to enable the parent platform to complete all... The mathematical model for the access sequence with the minimum total time for UUV replacement tasks is as follows:

[0096]

[0097]

[0098]

[0099] In the formula, Indicates access sequence The corresponding total replacement task time on the parent platform; Indicates the first Time taken for each pursuit segment; Indicates that the parent platform executes the first The relative still water propulsion speed vector required for each pursuit segment; for The set of all valid permutations of the above;

[0100] S16, the access sequence for each individual in the population. The earliest feasible rendezvous time is extracted segment by segment according to equation (6). The time spent on each flight segment Accumulate the total time consumed in the sequence and update the position of the parent platform segment by segment. ,in, This indicates that the parent platform has completed the first... The new position vector after replacing the target UUV; if there is no feasible rendezvous time in a certain flight segment throughout the entire time range, it is determined that a kinematic deadlock has been triggered, the recursion is immediately terminated and an extreme value penalty is applied, and it is excluded from the valid candidate set;

[0101] S2. Calculate the population fitness and dynamically monitor its evolutionary state, which includes the following sub-steps:

[0102] S21. Minimum task time for extracting valid individuals in the current generation. Compared with the average time of the population Define the evolutionary stagnation index:

[0103]

[0104] In the context of ocean currents, Dynamics reflect the diversity of effective paths. When When the value is relatively large, it indicates that there are still multiple differentiated mother platform replacement paths within the population, indicating an active macroscopic exploration period; when When the value is small, it indicates that the population has become highly homogeneous, and a large number of candidate replacement paths converge to similar local optimal paths. The access order of the parent platform tends to be similar, and it faces a serious risk of premature stagnation.

[0105] S22. Let the global historical best time consumption be... The previously recorded globally optimal time was ,like That is, the current generation discovers a shorter path to replace the parent platform, Update to the current optimal anti-flow field path. Continuous stagnation counter. The update formula is defined as follows:

[0106]

[0107] In the formula, By introducing a tolerance, the system can effectively filter out false advances caused by minor ocean current disturbances. Only when a current-resistant path that significantly shortens the total replacement time of the parent platform is discovered is the evolution considered valid and the counter reset. If the number of cases continues to accumulate and reaches a preset threshold, it is determined that the replacement path optimization of the parent platform has stalled, thereby triggering a switch between macro and micro mechanisms.

[0108] S3. Enter the adaptive genetics stage, dynamically adjusting the crossover and mutation probabilities of individuals, specifically including the following sub-steps:

[0109] S31. Select the generation with the shortest total time through a tournament to enter the parent pool; use a non-linear adaptive crossover probability based on individual fitness during the crossover phase. Let the time taken by the better parent among the two parent generations participating in the crossover be... The calculation rules are as follows:

[0110]

[0111] In the formula, and These are the upper and lower bounds of the crossover probability, respectively. When the fitness of the parent individual is better than the population mean, i.e. When the quality of the parent individual is lower than the population average, it indicates that the individual carries a gene segment with greater resistance to the flow field. At this point, the crossover probability decreases as the relative superiority of the individual increases, thus strictly protecting the elite downstream path and preventing the destruction of superior genes. This indicates that its gene sequence is poorly adapted to the current flow field, and it should be encouraged to recombine fully with other individuals; therefore, it is assigned the maximum crossover probability. This mechanism promotes genetic diversity. Essentially, it implements a differentiated genetic strategy of protecting superior genes and recombinating inferior ones, effectively avoiding the indiscriminate destruction of high-quality anti-fluid gene fragments by a fixed crossover probability.

[0112] S32. In the mutation phase, an exponential adaptive mutation probability based on the population stagnation index is adopted. :

[0113]

[0114] In the formula, and These are the upper and lower bounds of the mutation probability, respectively. The evolutionary stagnation index, It is a sensitivity modulator for variation, used to control Follow Sensitivity to change. When population diversity is sufficient... When fully large, trending towards the lower bound To avoid subjecting the normal evolutionary process to unnecessary random interference; when the population stagnates... When approaching zero, tending towards the upper bound It actively stimulates stronger mutation search, helping to break out of the stagnation zone;

[0115] S33. For each individual to be mutated, firstly, extract the time consumption data for each segment of its path, locate the upstream bottleneck segment where the mother platform approaches the UUV against the ocean current and has the highest time consumption, and then perform a flip operation on the local path segment centered on this segment, forcibly removing the high-time-consuming upstream segment from the mother platform's current access sequence. Compared to random flipping, this directional strategy can eliminate the risk of kinematic deadlock at the path level of ocean current interference in a more accurate way, and generate downstream candidate paths with lower resistance;

[0116] S4. Perform simulated annealing collaborative refining and deep stagnation processing, periodically refine the elite access sequence locally, and trigger deep simulated annealing and population reconstruction when the population is in deep stagnation. This includes the following sub-steps:

[0117] S41, When the continuous stop counter Reaching the set threshold At that time, a macro-micro switch is triggered to extract the current globally optimal access sequence. Deep simulation annealing is invoked as an initial solution to mitigate the impact. At each temperature, a neighborhood perturbation operation is randomly applied to the current parent platform access sequence to generate new candidate solutions. The neighborhood perturbation employs two alternating operations: first, a 2-opt flip, randomly selecting two positions in the sequence and flipping the UUV access subsequence between them; second, a node swap, randomly selecting two UUVs and swapping their positions in the access sequence. The acceptance of candidate solutions follows the Metropolis criterion.

[0118]

[0119] In the formula, This indicates the current temperature during the simulated annealing process. This represents the acceptance probability of a candidate access sequence. This represents the difference in total time taken between the candidate solution and the current solution for replacing the parent platform path. When When a candidate solution is superior to the current solution, it is accepted unconditionally; when... At that time, it accepts inferior solutions with probability, thus giving it the ability to escape the current local optimum. With temperature... According to the annealing coefficient As cooling progresses, the probability of accepting inferior solutions decreases continuously, and the search gradually converges from large-scale exploration to fine-grained local development. Deep annealing continuously records the optimal solutions found throughout the process. If a faster flow field-resistant path is found to replace the parent platform after annealing, the global optimal record is immediately updated.

[0120] S42. After deep annealing, regardless of whether improvement is achieved, a gradient blood replacement operation is performed: the first position in the population is forcibly filled with the current globally optimal access sequence. ; 2nd to 3rd The next best elite individuals are filled in according to their fitness ranking. The number of elite individuals retained during the tiered blood transfusion process is determined by the elite retention ratio. With population size determined; the first To the Replace all bits with a randomly generated valid access sequence. This refers to the population size. The counter will remain continuously idle after the blood transfusion is complete. Zeroing out. Furthermore, during the normal evolution phase, every [period]... The generation replaces some individuals with lower fitness rankings in the population with randomly generated legal access sequences according to a random injection ratio, as a normal supplement to suppress homogenization;

[0121] S43. When deep stagnation is not triggered, and the current evolution generation is within the embedded refinement trigger cycle. When the value is an integer multiple of the target value, initiate embedded collaborative refining: Select The optimal parent platform replaces the UUV path to perform local refinement, and the dynamic initial temperature is calculated for each elite individual. :

[0122]

[0123] In the formula, For temperature coefficient, The total time spent replacing the path for the current elite. With continuous optimization, Gradually decrease, Automatic temperature reduction allows the simulated annealing algorithm to automatically switch to a more refined local refining mode in the later stages, preventing the high-quality downstream path accumulated due to excessively high temperatures; while in the early stages, the larger... Give the simulated annealing algorithm a higher initial temperature With a stronger ability to escape, it helps to escape from early local suboptimal regions. This adaptive temperature design allows the search range of the simulated annealing algorithm to naturally shrink as the optimization process progresses, achieving a dynamic control effect of coarse adjustment in the early stage and fine adjustment in the later stage;

[0124] S44. Repeat the above steps until the maximum number of generations is reached. The iteration terminates, and the optimal planning path and shortest total time for the parent platform to perform dynamic collaborative replacement tasks for multiple UUVs are output.

[0125] To verify the effectiveness of the path planning method proposed in this invention, simulation verification was performed using MATLAB.

[0126] The parameters for this method are set as shown in Table 1:

[0127] Table 1 Parameter Data Table

[0128] Parameter name numerical values unit Population size 50 - Maximum number of generations 200 generation Cross upper bound 0.95 - Cross lower bound 0.60 - Upper bound of mutation 0.40 - Mutation lower bound 0.10 - Variation sensitivity modulator 0.10 - Deep simulation of initial annealing temperature 5 - Termination temperature 0.10 - Annealing attenuation coefficient 0.95 - Number of iterations 80 Second-rate Embedded refining annealing attenuation coefficient 0.96 - Embedded Refinement Iteration Steps 80 Second-rate Elite retention rate 0.20 - Embedded simulated annealing trigger cycle 10 generation Dynamic initial temperature coefficient 0.15 - Deep stagnation threshold 15 - Diversity Injection Cycle 50 generation

[0129] Following the parameter settings described above, the mother platform replaced 25 UUVs under a 0.5 knot ocean current environment, and simulations were performed using MATLAB. This method was compared and validated with traditional genetic algorithms and traditional simulated annealing algorithms. To ensure the statistical reliability and reproducibility of the results, 10 sets of random maps were independently generated. Each map used a fixed random seed to control the spatial layout and initial phase of the UUVs. All comparison methods on the same map were run under identical initial conditions. Finally, the numerical values ​​and mean values ​​of the 10 maps were used as the evaluation results for this scenario, as shown in Table 2.

[0130] Table 2 Simulation Results

[0131] map Genetic algorithm time (h) Simulated annealing algorithm time (h) The invention took (h) 1 52.40 43.77 39.60 2 47.83 45.74 39.96 3 52.01 37.25 34.11 4 61.28 43.33 41.65 5 53.97 40.53 37.82 6 52.61 43.68 40.79 7 58.96 44.20 36.75 8 55.93 43.92 41.06 9 51.30 39.22 37.41 10 47.80 40.62 37.91 mean 53.41 42.23 38.71

[0132] As shown in Table 2, the average time taken by the present invention on 10 maps is 38.71 hours, while the average time taken by the traditional genetic algorithm on 10 maps is 53.41 hours, and the average time taken by the traditional simulated annealing algorithm on 10 maps is 42.23 hours. The optimal path obtained by the present invention reduces the time taken by 27.52% compared with the traditional genetic algorithm and by 8.33% compared with the traditional simulated annealing algorithm.

[0133] Figure 2 This is the optimal replacement path for the simulated UUV in this invention. Figure 3 To simulate the optimal replacement path for UUVs using traditional genetic algorithms, Figure 4The optimal replacement path for UUVs was simulated using the traditional simulated annealing algorithm. Under the same scenario, the access paths obtained by the three algorithms showed significant differences. The genetic algorithm path contained many long-distance cross-regional segments and round-trip maneuvers, with noticeable local path intersections. The simulated annealing algorithm reduced some redundant segments through local neighborhood search, resulting in a more compact overall path than the genetic algorithm, but still exhibited long-distance crossings in the upper region and local detours. In contrast, the access path obtained by the method of this invention showed more obvious partition continuity, with tighter connections between adjacent access nodes, reducing long-distance counter-current approaches and repeated cross-regional voyages under unfavorable ocean current conditions, enabling the mother platform to complete the dynamic replacement of each UUV in a more reasonable order.

[0134] Figure 5 This is the cumulative replacement success curve of the present invention. Figure 6 This is the cumulative replacement success curve of a traditional genetic algorithm. Figure 7 The graph shows the cumulative replacement success rate of the traditional simulated annealing algorithm. The cumulative replacement count of all three algorithms increases over time, but their efficiency differs significantly. The method of this invention completes the replacement of all 25 UUVs in approximately 40 hours; the simulated annealing algorithm is next, completing it in approximately 43 hours; the genetic algorithm is the slowest, completing it in approximately 54 hours. This indicates that the method of this invention can complete the continuous replacement task faster, and its overall planning efficiency is superior to the two comparative algorithms.

[0135] Simulation results show that by introducing kinematic deadlock filtering, adaptive dual-operator genetics, and embedded simulated annealing collaborative refining mechanism, the present invention can effectively improve the replacement order of multiple UUVs by the parent platform and reduce the overall replacement task time, thus verifying the effectiveness and feasibility of the method of the present invention in dynamic replacement path planning of multiple UUVs in ocean current environment.

[0136] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the technical solutions of the present invention have been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the protection scope of the present invention.

Claims

1. An adaptive dual-operator and embedded refined multi-UUV replacement path planning method, characterized in that, Includes the following steps: S1. Establish a system kinematic model of the parent platform and multiple UUVs, construct the dynamic cooperative replacement task as a moving target traveling salesman problem, and perform deadlock filtering. This includes the following sub-steps: S11. Establish a global two-dimensional rectangular coordinate system Assume there are [illegible] sea areas. Let the set of UUVs to be replaced be . ,in, For UUV numbering, Indicates the first One UUV to be replaced; for any UUV Let the coordinates of the geometric center of its circumference be... ,in, and They represent the first The geometric center of the UUV cruising trajectory is at shaft and The coordinates along the axis, the travel radius is With constant waterline velocity Conduct a circular tour; S12. Assume the current vector in the operating sea area is constant. ,in, and These represent the ocean current speeds at... shaft and Components in the axial direction; let This represents the actual velocity vector of the parent platform relative to the ground. The relative hydrostatic thrust velocity vector of the parent platform; The maximum powered speed of the mother platform in still water conditions, and the ground motion of the mother platform satisfies the Zermelo navigation equations: S13. To ensure that the rendezvous is physically feasible, the hydrostatic thrust required by the mother platform must strictly satisfy the kinematic boundary constraints: when If the condition is always true, then the access sequence is determined to be physically unreachable. S14. Let the access sequence of the parent platform be... ,in, This represents the initial node of the parent platform. For sequences... The first in For each pursuit segment, determine the earliest feasible rendezvous time that satisfies the dynamic boundary conditions: In the formula, Indicates the moment when candidates meet. Indicates the previous meeting point. Indicates the first The earliest feasible rendezvous time for each leg of the journey. Indicates the parent platform in the Position vector at the start of each flight segment Indicates the target UUV at time... The dynamic position vector; If equation (3) has no solution over the entire time interval, an extreme value penalty is imposed on its fitness: , Fitness of the access sequence; S15. Define the path planning problem as follows: under the conditions of satisfying Zermelo kinematic constraints and each UUV being visited exactly once, solve the problem to enable the parent platform to complete all... The mathematical model for the access sequence with the minimum total time for UUV replacement tasks is as follows: In the formula, Indicates access sequence The corresponding total replacement task time on the parent platform; Indicates the first Time taken for each pursuit segment; Indicates that the parent platform executes the first The relative still water propulsion speed vector required for each pursuit segment; for The set of all valid permutations of the above; S16, the access sequence for each individual in the population. The earliest feasible rendezvous time is extracted segment by segment according to equation (3). The time spent on each flight segment Accumulate the total time consumed in the sequence and update the position of the parent platform segment by segment. ,in, This indicates that the parent platform has completed the first... The new position vector after replacing the target UUV; if a kinematic deadlock is triggered, the recursion is immediately terminated and an extreme value penalty is applied; S2. Calculate the population fitness and dynamically monitor its evolutionary state, which includes the following sub-steps: S21. Minimum task time for extracting valid individuals in the current generation. Compared with the average time of the population Define the evolutionary stagnation index: S22. Let the global historical best time consumption be... The previously recorded globally optimal time was Continuous stall counter The update formula is defined as follows: In the formula, The tolerance is set; S3. Enter the adaptive genetics stage, dynamically adjusting the crossover and mutation probabilities of individuals, specifically including the following sub-steps: S31. Select the generation with the shortest total time through a tournament to enter the parent pool; use a non-linear adaptive crossover probability based on individual fitness during the crossover phase. Let the time taken by the better parent among the two parent generations participating in the crossover be... The calculation rules are as follows: In the formula, and These are the upper and lower bounds of the crossover probability, respectively. This represents the average time consumed by the population. S32. In the mutation phase, an exponential adaptive mutation probability based on the population stagnation index is adopted. : In the formula, and These are the upper and lower bounds of the mutation probability, respectively. As a regulator of sensitivity to variation, It is an index of evolutionary stagnation. S33, Perform reverse flow directional reversal; S4. Perform simulated annealing, co-refining, and deep stagnation processing, which includes the following sub-steps: S41, When the continuous stop counter Reaching the set threshold At that time, a macro-micro switch is triggered to extract the current globally optimal access sequence. The deep simulated annealing algorithm is invoked as the initial solution; S42. After deep annealing, perform gradient blood exchange operation; S43. When depth stagnation is not triggered and it is within the embedded refining trigger cycle. When the temperature is an integer multiple of the target value, initiate embedded genetic-simulated annealing co-refinement: select the optimal path to perform local refinement, and calculate the dynamic initial temperature for each elite individual. : In the formula, For temperature coefficient, Total time spent replacing the path for the current elite; S44. Repeat the above steps until the maximum number of generations is reached. The iteration is terminated.

Citation Information

Patent Citations

  • Improved GA-SA path planning method of unmanned surface vehicle for offshore wind plant inspection

    CN111612217A