LU101833B1

LU101833B1 - Route Planning Method and System for Unmanned Surface Vehicles Using Particle Swarm Optimization Based on Greedy Mechanism

Info

Publication number: LU101833B1
Application number: LU101833A
Authority: LU
Inventors: Junfeng Xin; Shixin Li; Penghao Li; Fengru Yang; Shuyue Li
Original assignee: Qingdao Blue Ocean Future Tech Co Ltd
Priority date: 2020-03-10
Filing date: 2020-06-05
Publication date: 2020-10-05
Also published as: CN111338350A

Abstract

The present invention belongs to the route planning field, providing a kind of route planning method and system for USVs using the particle swarm optimization based on greedy mechanism. It solves the problem of low efficiency of USV route planning, and realizes the fast USV route planning. The USV route planning method includes obtaining the current position, course data and target position of the USV; calculating the optimal position for the USV to sail from the current position to the target position by the particle swarm optimization based on greedy mechanism, and then modifying the trajectory deflection of the USVs course data according to the preset constraint factor, and finally getting the optimal route of the USV.

Description

lu101833 Route Planning Method and System for Unmanned Surface Vehicles Using Particle Swarm Optimization Based on Greedy Mechanism Field of Technology The present invention belongs to the route planning field, especially involving a kind of route planning method and system for unmanned surface vehicles using the particle swarm optimization based on greedy mechanism. Background of the invention The statements in this section only involve the related background information of the present invention, and do not necessarily constitute the prior art.

Unmanned surface vehicles (USVs), also known as autonomous surface vehicles, have aroused wide concern all over the world in the commercial, scientific and military fields in recent years. USVs have the advantages of low operation and maintenance costs, reducing risk of casualties, and good maneuverability and reliability under different operating conditions. By means of effective and reliable navigation equipment such as global positioning system (GPS), wireless communication units and various types of sensors, USVs can be cost-effectively used in applications, including underwater survey, pollutant tracking, acoustic navigation, marine rescue, and obstacle detection. Great progress has been made in the related technologies of the USV system in recent years, but the improvement of the autonomy of USV in complicated or dangerous situations is still a challenge. Further development is required for key issues such as hull hydrodynamics, communication technology, and navigation, guidance and control (NGC) strategy. Route planning is a significant part of the USV navigation system. It has great significance for the design of feasible and optimal trajectories of navigation-based control systems, as well as for update information, mission needs, and environmental conditions. Its effectiveness not only determines the autonomy of USVs, but also affects the reliability and efficiency of mission execution. The USV route planning problem is usually described as a traveling salesman problem (TSP), i.e. a typical combinational optimization problem. Specifically, the problem is to find a shortest closed loop that passes through all target cities without repeat.

lu101833 The inventor found that the number of possible routes increases exponentially as the number of target points increases when the USV executes multi-target missions in a complicated marine environment, resulting in the so-called exponential explosion. In this case, traditional algorithms such as method of exhaustion and branch and bound method cannot find the optimal solution within a reasonable time cost, leading to slow and inefficient route planning for USVs. Summary of the invention In order to solve the above problems, the first aspect of the present invention provides a route planning method for USVs using the particle swarm optimization based on greedy mechanism. Based on a greedy black box, this method ensures the particles to constantly move towards higher fitness, and maintains the population diversity to a certain extent. Besides, this method uses the 2-opt operation to retain excellent route segments from the old solution and eliminate route crossover, greatly reducing the time for obtaining an optimal USV route planning.

For the above mentioned purposes, the present invention is provided with the following technical scheme.

A route planning method for USVs using particle swarm optimization based on greedy mechanism, which includes: obtaining the current position, course data and target position of the USV; calculating the optimal position for the USV to sail from the current position to the target position by the particle swarm optimization based on greedy mechanism, and then modifying the trajectory deflection of the USV's course data according to the preset constraint factor, and finally getting the optimal route of the USV.

In particular, the process of calculating the optimal position for the USV to sail from the current position to the target position by the particle swarm optimization based on greedy mechanism: - 25 establishing a greedy black box, initialize the particle swarm parameters based on the position of the USV, and generating an initial particle swarm; the initial particle swarm is composed of several initial routes of the USV, and the particle is the location point on the route;

lu101833 enteringthe iterative loop, and dividing particles in the initial particle swarm of each iteration into preset number of particle groups; screeningout the two particles O (1,2) with the lowest fitness in each particle group, and generating two new particles O' (1,2) using the greedy black box; the fitness is the reciprocal of the route length; comparingthe fitness of O (1,2) and O' (1,2) to keep the two particles with high fitness and update the particle group; performing the 2-opt operation on all updated particle groups and finding the shortest route in each updated particle group, updating the optimal particle position and the optimal group position until the conditions for stopping the iteration loop are reached, and outputting the current optimal group position.

In order to solve the above problems, the second aspect of the present invention provides a route planning system for USVs using the particle swarm optimization based on greedy mechanism. Based on a greedy black box, this method ensures the particles to constantly move towards higher fitness, and maintains the population diversity to a certain extent. Besides, this method uses the 2-opt operation to retain excellent route segments from the old solution and eliminate route crossover, greatly reducing the time for obtaining an optimal USV route planning.

A route planning system for USVs using the particle swarm optimization based on greedy mechanism, including: data acquisition module, which is used to acquire the current position, course data and target position of the USV; route planning module, which is used to calculate the optimal position for the USV to sail from the current position to the target position by the particle swarm optimization based on greedy mechanism, and then modify the trajectory deflection of the USV's course data according to the preset constraint factor, and finally get the optimal route of the USV; in particular, the process of calculating the optimal position for the USV to sail from the current position to the target position by the particle swarm optimization based on greedylu101833 mechanism: establishing a greedy black box, initialize the particle swarm parameters based on the position of the USV, and generating an initial particle swarm; the initial particle swarm is composed of several initial routes of the USV, and the particle is the location point on the route; entering the iterative loop, and dividing particles in the initial particle swarm of each iteration into preset number of particle groups; screening out the two particles O (1,2) with the lowest fitness in each particle group, and generating two new particles O' (1,2) using the greedy black box; The fitness is the reciprocal of the | route length; comparing the fitness of O (1,2) and O' (1,2) to keep the two particles with high fitness and update the particle group; performing the 2-opt operation on all updated particle groups and finding the shortest route in each updated particle group, updating the optimal particle position and the optimal group position until the conditions for stopping the iteration loop are reached, and outputting the current optimal group position.

The third aspect of the present invention provides a computer-based readable memory medium that stores a computer program. When the program is executed by the processor, the steps in the said route planning method for USVs using the particle swarm optimization based on greedy mechanism described above are realized.

The fourth aspect of the present invention provides a computer equipment, including a memory, a processor, and a computer program that are stored on the memory and can be executed on the processor. When the said processor executes the said program, the steps in the said route planning method for USVs using the particle swarm optimization based on greedy mechanism described above are realized.

Beneficial effects of the invention are: (1) In the present invention, a greedy black box is used to initialize the particles, which overcomes the randomness of the traditional method, excludes some infeasible solutions at the beginning of optimization, and improves the efficiency of optimizing USV's routes.

(2) In the present invention, a greedy black box is used to ensure particles move constantly towards a higher degree of fitness, and maintain the swarm diversity to a certain extent.

lu101833 Furthermore, the 2-opt operation can retain excellent encoded string segments from the old solution and eliminate route crossover, thus greatly reducing the time spent on the route planning of USVs. Under the premise of ensuring the quality of the solution, the robustness of the algorithm is greatly improved, and the optimal route length is sharply shortened, thus achieving the purpose of rapidly 5 realizing the optimal route planning for USVs.

(3) The present invention also corrects the track deflection of the current course data according to the constraint factors. Thus, the excessive deviation of the planned route from the actual route caused by the track deflection error is avoided, and the accuracy of planned routes of USVs is | improved.

BRIEF DESCRIPTION OF THE DRAWINGS Drawings forming part of this invention are used to provide further understanding herein. The illustrative embodiments of this invention and their explanations are used to interpret this invention and do not constitute undue restrictions herein.

Fig. 1 is the flow diagram of a route planning method for USVs using the particle swarm optimization based on greedy mechanism for the embodiment in the present invention.

Fig. 2 is the initial travelling route of the embodiments in the present invention.

Fig. 3 is the 2-opt operation mechanism of the embodiments in the present invention.

Fig. 4 shows a comparison of classical genetic algorithm (CGA), ant colony optimization (ACO), classical particle swarm optimization (CPSO), and improved particle swarm optimization (IPSO) of the embodiments in the present invention.

Fig. 5 (a) through Fig. 5 (h) show the results of comparison of classical genetic algorithm (CGA), ant colony optimization (ACO), classical particle swarm optimization (CPSO), and particle swarm optimization based on greedy mechanism (IPSO) when the maximum number of iterations is 300, 400, 500, 600, 700, 800, 900 and 1000.

Fig. 6 (a) through Fig. 6 (h) are the iterative history of the optimal route distance (D) and iteration (m) corresponding to Fig. 5 (a) through Fig. 5 (h). Fig. 7 (a) through Fig. 7 (h) show the optimal trajectories generated by the eight TSPLIBlu101833 instances of the embodiments in the present invention based on the ant colony optimization (ACO).

Fig. 8 (a) through Fig. 8 (h) show the optimal trajectories generated by the eight TSPLIB instances of the embodiments in the present invention based on the improved particle swarm optimization (IPSO).

Fig. 9 is the structural representation of the route planning system for USVs using the particle swarm optimization based on greedy mechanism of the embodiments in the present invention. Detailed description of the Embodiments The present invention is further described in connection with the drawings and embodiments. It is important to note that the following detailed description is illustrative and aims to provide a further description of this invention. Unless otherwise indicated, all technical and scientific terms used herein have the same meaning as commonly understood by ordinary technical personnel in the field of technology to which present invention belongs.

It should be noted that the terms used herein is to describe the specific mode of execution, but not to impose restrictions on the exemplary mode of execution of this invention. Unless otherwise expressly stated, the singular form is also intended to include the plural form. In addition, it should be understood that the terms "contain" and/or "include" used in this Specification indicate the presence of features, steps, operations, devices, components and/or combinations of them, Embodiment 1, the present embodiment provides a route planning method for USVs using the particle swarm optimization based on greedy mechanism, including: Step I: obtaining the current position, course data and target position of the USV; Step II: calculating the optimal position for the USV to sail from the current position to the target position by the particle swarm optimization based on greedy mechanism, and then modifying the trajectory deflection of the USV's course data according to the preset constraint factor, and finally getting the optimal route of the USV. In practice, the process of calculating the optimal position for the USV to sail from the currentlu101833 position to the target position by the particle swarm optimization based on greedy mechanism: Step 1: Establishing a greedy black box, initialize the particle swarm parameters based on the position of the USV, and generating an initial particle swarm; the initial particle swarm is composed of several initial routes of the USV, and the particle is the location point on the route; The initialized particle swarm parameters include the number of initialized populations, the | maximum number of iterations, the calculation function of fitness, and the velocity correlation coefficient of particles. The calculation function of fitness is the reciprocal of the route length.

The technical difficulties of combining the greedy mechanism with the particle swarm optimization in this embodiment: when the greedy mechanism generates particles through the particle black box, it is required to first determine the first traversal city of the particle to be replaced. However, during the iteration, the first traversal city will change, as the encoding of particles in the optimizing phase of the particle swarm itself change to some extent. Therefore, the first traversal city number of particles to be planned must be determined in real time while performing the greedy mechanism. Only in this way can new particles be generated through the particle black box.

The principle of the greedy algorithm is often described as making a selection that currently appears to be locally optimal, and intending to find a satisfactory solution within a reasonable time. Previous decisions can be taken as references when making such a selection, but the future selections or other inherent selections of sub-problems should not be depended on. In other words, the greedy algorithm is short-sighted and never reconsiders its previous decisions. This is completely different from dynamic planning. The dynamic planning will reconsider previous decisions to find the optimal solution after exhaust operations. Furthermore, most of the greedy algorithms cannot achieve the global optimum. However, it is still widely used as an auxiliary strategy or to give the optimal approximation to some problems of time constraints. Moreover, the greedy algorithm is also effective for the optimal substructure problem in which the globally optimal solution includes the locally optimal solution of sub-problem. The greedy algorithm is used when solving the traveling salesman problem (TSP). The nearest city that has not been visited is selected as the next target of the salesperson until a completely closed route is generated. It is usually time-consuming to get the optimal solution of the traveling salesman problem (TSP). Thus,

lu101833the greedy algorithm has been integrated into many other algorithms in recent years to velocity up the rate of convergence of the algorithm and improve the search efficiency.

Pan et al. proposed a hybrid algorithm that has eliminated route crossover and delete crossover operators.

In this algorithm, the global search uses an immune algorithm, and the population initialization uses a greedy algorithm.

The results show that this method improves the reliability, rate of global convergence and search capability of the immune algorithm.

In order to reduce the computation time under the premise of ensuring the quality of the solution, Basu et al. adopted the stochastic greedy contraction method for the preprocessing of image sparsity before using tabu search to solve the asymmetric TSP.

This method is compared with other successful heuristic methods, finding that this method will reduce the time cost by 1% -5%. Combining greedy random adaptive search process (GRASP), iterated local search and iterated variable neighborhood descent, Mestria proposes a hybrid clustering TSP method.

The hybrid heuristic method is superior to the existingoptimal methods and the reasonable methods for calculating time for medium and large instances.

Inspired by bird flight models, the particle swarm optimization (PSO) was proposed by Eberhart and Kennedy in 1995. In order to avoid the collision of real bird flocks, the bird flocks are abstracted into random particle swarms without mass and volume parameters to obtain a globally optimal solution.

Every particle moves in the search space under the dynamic guidance of its own flying status, own experience, and group experience.

A fitness function is used to evaluate the solution, expecting the most satisfactory solution.

Moreover, particle swarm optimization has the advantages of fast convergence, simple parameter setting, and easy realization, which make it be widely used in a large number of discrete and continuous optimization such as route planning problems, molecular docking problems, electrochemical machining and image recognition, as well as semi-ideal facility location problems and sheet metal forming.

To improve the effectiveness of particle swarm optimization in solving TSP, a lot of valuable research on the combination of two or three heuristic algorithms have been conducted.

Shuang et al. improved the ant colony optimization (ACO), introduced PSO to expand the search space, and used group experience to accelerate convergence.

Mahi et al. also verified similar strategies that introduce PSO into the ant colony optimization to optimize city selection parameters.

Zhang et al. proposed another improved particle swarm optimization, which adopts a priority encoding method to encode the solution vector, dynamically sets the range of velocity, eliminates the side effects of the discrete search space, andlu101833 uses the k-centering method to avoid falling into local optimum. This combinational algorithm has good performance in retaining group diversity. In the study of Feng et al, the adaptive fuzzy C-means algorithm was used to divide the large-scale search space into subspaces. Then, the | particle swarm optimization based on transformation was combined with the simulated annealing algorithm to find the locally optimal solution. Finally, a complete route was reconstructed using the max-min merge algorithm. Experiments on the combination of particle swarm optimization with genetic algorithm and artificial fish swarm algorithm have also achieved satisfactory results.

Particle swarm optimization based on greedy mechanism: On the one hand, particle swarm optimization is easy to fall into local optimum and is not effective in solving discrete problems, which are its inherent limitations. On the other hand, the principle of the greedy algorithm is relatively simple, easy to conduct with high efficiency, although a global optimal solution may not necessarily be obtained. Thus, it seems completely feasible to combine these two algorithms to solve various optimization problems utilizing their complementary advantages.

The traditional particle swarm optimization randomly spreads N particles in the S-dimensional search space at the beginning of the evolutionary process. The position and velocity of the i-th particle can be represented by vectors Xi = (xi1, Xi2, …Xis)" and Vi = (vi1, Viz, ..., Vis)”, respectively. The fitness function is defined as 1/D; (Di means the route length), and the personal best position (Pis) and group best position (Pg) are updated in each iteration process. At the same time, the velocity and position of each particle are updated as: vist = wo + OUPS — xi) + cor (FE — x), (1 Xie = aff + vt 2) Wherein, m represents the current number of iterations, and s represents the s-th dimension. riand r2 are iteratively updated random numbers that evenly distribute between 0 and 1. c1 and c2 are individual cognitive coefficient and social cognitive coefficient, usually expressed by 2. w is the inertia weight. Algorithm 1 describes the pseudo code of traditional particle swarm optimization.

What calls for special attention is that the PSO control parameters determine the balance degree between exploration (searching for a wider space) and development (moving to a local

| 10 . lu101833 optimum), which has a great influence on the algorithm performance.

For example, the acceleration coefficients cı and cz play a vital role in balancing the impact of individual cognition and social cognition on guiding particles towards the optimal solution.

The stochastic behavior of ri andr; can maintain the population diversity to a certain extent, and avoid premature convergence. — Algorithm 1: Traditional particle swarm optimization solves the route planning problem 1 Select the number of populations and the maximum number of iterations 2 Determine the fitness function 3 Set individual cognitive coefficient c1, social cognitive coefficient c2, and inertia weight w 4 For each particle do Initialize velocity and position 6 Evaluate the fitness function 7 Record the initial personal best position (Pis) and group best position (Pgs) 8 End 9 While when the maximum number of iterations or the minimum error does not reach the standard do For each particle do 11 Update the velocity by formula (1) 12 Update the position by formula (2) 13 Calculate the new fitness function 14 Update the initial personal best position (Pis) and group best position (Pgs) End 16 End 5 It is well known that the traditional particle swarm optimization generates the initial particle swarm randomly.

This will produce some infeasible solutions, thus limiting the rate of convergence and search efficiency.

Thus, the greedy mechanism and the 2-opt operation are integrated into PSO, expecting to improve the effectiveness of the algorithm.

In our new algorithm, a greedy black box is used in the particle initialization and particle generation stages of each iteration.

However, the 10 greedy mechanism only considers the current local search situation, which make it easy to generatelu101833 route crossover.

Consequently, it is necessary to use the 2-opt operation to increase the probability | of generating a satisfactory combination between locally optimal segments, and to eliminate route crossover.

The cities on the route are taken as an example: A greedy black box is established based on the greedy mechanism.

It has the function of establishing a locally optimal solution after determining the departure city.

For the TSPs of N cities, C cities closest to the i-th city can be found through distance measurement.

Therefore, a matrix An x c with the size of NxC is defined.

Ajj is the city code, meaning the city in the j-th city that is closest to the i-th city.

A [i] is the code of C cities nearest to the i-th city.

In this study, C is set as 3. Taking the burmal4 in TSPLIB as an example, the mechanism of the greedy black box can be described as follows.

If the salesman starts from the third city, the code 14 is selected from A [3] as the next city to visit.

Then it can be known from A [14] that the twelfth city is the nearest city and should be selected as the next city to visit.

The next city to visit is the sixth city, as shown in A [12]. By parity of reasoning, we will select the following cities to form a travelling route.

In this process, checks are performed to avoid repeat visit.

Generally, three columns can meet the requirements of generating new solutions.

However, if repetition cannot be avoided, the next city to visit can be selected randomly for repeatability check until a closed route is formed.

In this principle, the initial travelling route can be represented by a series of city codes L = {3, 14, 12, 6, 7, 13, 8, 1, 11, 9, 10, 2, 5, 4}, as shown in Fig. 2. The matrix A14 x 3 is shown as below: 8 11 9 1 8 11 14 4 12 12 3 14 6 12 7 12 7 14 _ 113 12 6 3 A14x3 — 1 11 9 ( ) 11 1 8 9 11 1 9 8 1 6 14 7 7 8 12 12 3 6

| lu101833 For one thing, the greedy black box is used to eliminate a large number of infeasible solutions caused by the randomness of the traditional method, resulting in high-quality initial particle swarms. For another, the greedy black box strategy makes the newly planned routes composed of multiple locally optimal segments, featuring short distance, low order, and well adaptability. As a result, it is feasible and reasonable to use the greedy black box for particle initialization.

Step 2: entering the iterative loop, and dividing particles in the initial particle swarm of each iteration into preset number of particle groups; Step 3: screening out the two particles O (1,2) with the lowest fitness in each particle group, and generating two new particles O' (1,2) using the greedy black box; Step 4: comparing the fitness of O (1,2) and O' (1,2) to keep the two particles with high fitness and update the particle group; Step 5: the 2-opt operation is performed on all updated particle groups to calculate the fitness between particles, and update the particle best position and the group best position. The current group best position is output when the stop conditions for iterative loop are reached, and then the track deflection of the current course data is corrected according to constraint factors. The optimal route for the USV is finally obtained.

The local search is used to improve the quality of the generated solutions. In the iteration process, all particles are scrambled and divided into 4 subgroups after the velocity and position of particles are updated. The "unordered" execution will avoid repeated replacement of the same particle as much as possible. Besides, the grouping operation helps establish a contact platform within a subgroup or between any two subgroups. Two particles with lower fitness values are selected from each subgroup. Their first city is used by the greedy black box to generate two new particles. The newly generated particles will replace the old particles if they have higher fitness values; otherwise, no replacement will be performed. This operation leads to the survival of the fittest to a certain extent, and prevents particle swarms from being completely generated by greedy black boxes, also known as greedy selection strategy. Furthermore, only locally optimal codes, selections, and non-repeatability are considered when greedy black boxes are used to generate particles, so some route crossover may occur, such as the intersection of the sixth city, as shown in

| 13 | lu101833 Fig. 2.

Therefore, the 2-opt operation is performed in each iteration to eliminate the route crossover. We expect to keep that the solution vector has locally optimal segments, and increase the probability of generating satisfactory combinations between these segments.

The 2-opt operation was first proposed by Croes in 1958. The structure is as shown in Fig. 3 The two sides are randomly deleted to divide the closed route into two parts. Then, the ends of the two parts are reconnected in another appropriate way to create a new solution. In other words, two non-adjacent city nodes are randomly selected from the perspective of route coding. The route segments between the two non-adjacent city nodes are completely reversed and connected back to the original coded strings, leading to a new solution. This process will be repeated until the shortest route is found.

Theoretically, the 2-opt operation ensures that the newly generated solution reserves the excellent coded string segments in the old solution. Moreover, the greedy selection strategy guarantees the unipolarity of the optimized solution vectors. In other words, the particles constantly move towards higher fitness.

Algorithm 2 gives the pseudo code of the improved particle swarm optimization. First, a greedy black box is established to generate the initial particle swarm that consists of multiple locally optimal solutions. In each iteration, the fitness of each particle will be calculated to update the individual and optimal solutions. Then, the velocity and position of each particle is updated by Eq. (1) and Eq. (2). Besides, the greedy selection strategy is applied to the local search, together with the 2-opt operation. When the maximum number of iterations is reached or a short enough route is found, the algorithm is terminated.

Algorithm 2: Improved particle swarm optimization solves the route planning problem 1 Select the number of populations and the maximum number of iterations 2 Determine the fitness function 3 Set individual cognitive coefficient c1, social cognitive coefficient c2, and inertia weight w

| lu101833 4 Set up a greedy black box for particle initialization While when the maximum number of iterations or the minimum error does not reach the standard do 6 Generate initial particles by traditional particle swarm optimization 7 Disorganize initial particles and subgroups 8 For each particle swarm do 9 Rank the fitness value and find 2 particles with lower fitness O (1,2) Generate 2 new particles O'(1,2)by the greedy black box Compare the fitness values of O (1,2) and O' (1,2), and reserve the 2 particles with higher fitness End 11 2-opt operation 12 Calculate the particle fitness 13 Update the individual optimum and global optimum 14 End The said constraint factors are converted from the weather and sea wave information of the real-time environment where the USV is located.

The weather and sea wave information of the environment where the USV is located is collected by the ultrasonic weather sensor, including the height, flow velocity and wavelength of the 5 sea wave. The weather and sea wave information are converted into constraint factors and applied to correct the trajectory of USVs.

The said constraint factors are mainly the sea wave force. Since the force acting on the vessel in water is mainly the sea wave force, the sea wave force model is used as the constraint factor to correct the track deflection. The function of constraint factors can be written as: Fo =-0.0073V;"hV; +0.0057v , My = = NEL 10 Wherein, h is the height of the sea wave; Vi is the velocity of the sea wave; Ao is thelu101833 wavelength of the sea wave; Mo is the mass of the sea wave.

The number of city points is usually selected based on experience to ensure that route crossover can be completely eliminated by the 2-opt. In addition, the preliminary relationship between the number of city points and the referenced maximum number of iterations was obtained through early tests, as shown in Fig. 4.

Classical genetic algorithm (CGA) and ant colony optimization (ACO) are compared with classical PSO (CPSO) and particle swarm optimization based on greedy mechanism (IPSO). Eight instances including eil51, rat99, kroal00, lin105, ch150, kroa200, tsp225, and lin318 from TSPLIB are used. To remove the randomness of the algorithm in the MATLAB® operating environment, we conducted 100 times of Monte Carlo simulation to obtain the data set of the optimal route distance (D) for each instance and algorithm. In addition, the control parameters of each algorithm are shown in Tab. 1 to facilitate the repeatability of this work. It is worth highlighting that the maximum number of iterations (max) depends on the number of city points in order to strike a balance between complete convergence and economic time consumption. This has been determined through preliminary tests. The maximum values of these 8 instances are 300, 400, 500, 600, 700, 800, 900, and 1000, respectively.

Tab. 1: Parameterization of CPSO, CGA, ACO and IPSO Algorithm Parameters Setting value Urban scale R 500 Individual cognitive coefficient cı Constant, 2

CPSO Social cognitive coefficient c2 Constant, 2 Initial weight w Constant, 0.9 5 Ban sale R 0 CGA Crossover probability Pc Constant, 0.9 Mutation probability Pm Constant, 0.1 Caco Ubmeainr 0lu101833 Information heuristic factor a Constant, 1 Expectation heuristic factor B Constant, 5 Pheromone evaporation rate p Constant, 0.2 Urban scale R 500 Individual cognitive coefficient cı Constant, 2

IPSO Social cognitive coefficient c2 Constant, 2 Initial weight w Constant, 0.9 Fig. 5 (a) through Fig. 5 (h) show the results of comparison of classical genetic algorithm (CGA), ant colony optimization (ACO), classical particle swarm optimization (CPSO), and improved particle swarm optimization (IPSO) when the maximum number of iterations is 300, 400, 500, 600, 700, 800, 900 and 1000. The comparison results are presented in the form of block diagrams and line graphs. Spear's explanation was mentioned. A box is drawn to represent the quartile range of the data set in each plot. It can reflect the degree of data dispersion to a certain extent, or the robustness of the algorithm. In addition, a red line and a plus sign are drawn in the bar diagram to identify the median and mean of the data set. The stick has whiskers on both sides, with their ends representing the optimal value and the worst value, respectively. The results of ACO and IPSO increase due to their small differences in the top right corner. In addition, Tab. 2 shows detailed information, including the form of the known optimal solution (KOS), average (AVG), standard deviation (SD), relative percentage error (RPE), and number of critical iteration (Meri). It should be noted that SD is calculated to show the distance between all data points (Dx) and the quantized average value of the algorithm robustness. Besides, RPE is defined to reflect the difference between the average solution and TSPLIB's KOS. In the four algorithms of these eight instances, the optimal value of AVG is shown in bold and highlighted in gray. SD and RPE are defined as: 1 SD = 25109 (D, — AVG)?, (4) (AVG — KOS) %)=-—— 5 RPE(%) KOS <100 (5)

lu101833 Generally speaking, both CPSO and CGA lead to the difference solution with a large AVG and a large degree of data dispersion in these eight instances. There is still much room for improvement for their effectiveness. By contrast, ACO and IPSO show similar and satisfactory performance, especially when more city points are considered. Through the observation of the amplifying images, IPSO can find a short route, despite of the relatively poor robustness. Taking lin105 as an example, CPSO gives an exaggerated solution with AVG as 60625m and SD as 2978m. But IPSO effectively reduces AVG by 74.9% and SD by 89.3% with the help of greedy mechanism and 2-opt. In terms of tsp225, the average optimal route distances of ACO and IPSO are 4300 m and 4188 m, respectively. The SD value of IPSO is only 0.6% greater than that of ACO. Besides, it is found that the difference between the IPSO's AVG and TSPLIB's KOS is less than 10% in all cases.

Besides, Fig. 6 (a) through Fig. 6 (h) are the iterative history of the optimal route distance (D) and iteration (m) corresponding to Fig. 5 (a) through Fig. 5 (h). On the whole, the curve of each algorithm shows a similar development trend. The value of D decreases obviously before reaching the number of critical iterations (Mc), as the number of iterations increases, and then the solution reaches convergence. This embodiment uses the Meni criterion to evaluate the rate of convergence and computing efficiency of each algorithm. The results show that CPSO converges quickly and terminates at the minimum Mori. Obviously, this will lead to a local optimum. Compared with the other three algorithms, CGA shows a very slow rate of convergence in all cases. This kind of behaviour helps to find a better solution than CPSO. Meanwhile, IPSO and ACO are consistent with each other in terms of rate of convergence and optimal route distance after convergence. Taking kroal00 as an example, CGA requires 446 times of iteration to achieve convergence, which is 7 times that of CPSO. The results indicate that the proposed algorithm can reduce the average optimal route distance by about 42.9%. Besides, the number of critical iterations of IPSO and CPSO are 113 and 62, respectively. Meri is increased by 82%, which reduces the AVG from 85243.19 to

22981.74m. It is necessary to properly slow down the rate of convergence to obtain a high-quality solution.

CPSO and CGA will present the chaotic routes of eight TSPLIB instances. Thus, Fig. 7 (a) through Fig. 7 (h) are the optimal trajectories generated by the eight TSPLIB instances of this embodiment based on the ant colony optimization (ACO); Fig. 8 (a) through Fig. 8 (h) show thelu101833 optimal trajectories generated by the eight TSPLIB instances of the embodiments in the present invention based on the improved particle swarm optimization (IPSO). The routes of ACO show different degrees of crossover, especially when more planning points are considered. However, the routes of IPSO show no crossover. The application of 2-opt operation and greedy selection strategy helps to avoid route crossover and reduce route complexity. This is also why ACO obtains a longer route than IPSO under the same conditions. The results show that this is of vital importance to improve the quality of solutions, avoid the local optimum, and delay the rate of convergence.

Embodiment 2, the present embodiment provides a route planning system for USVs using the particle swarm optimization based on greedy mechanism, including (1) data acquisition module, which is used to acquire the current position, course data and target position of the USV; (2) route planning module, which is used to calculate the optimal position for the USV to sail from the current position to the target position by the particle swarm optimization based on greedy mechanism, and then modify the trajectory deflection of the USV's course data according to the preset constraint factor, and finally get the optimal route of the USV.

In particular, the process of calculating the optimal position for the USV to sail from the current position to the target position by the particle swarm optimization based on greedy mechanism: establishing a greedy black box, initialize the particle swarm parameters based on the position of the USV, and generating an initial particle swarm; the initial particle swarm is composed of several initial routes of the USV, and the particle is the location point on the route; entering the iterative loop, and dividing particles in the initial particle swarm of each iteration into preset number of particle groups; screening out the two particles O (1,2) with the lowest fitness in each particle group, and generating two new particles O' (1,2) using the greedy black box; the fitness is the reciprocal of the route length;

lu101833 comparing the fitness of O (1,2) and O' (1,2) to keep the two particles with high fitness and update the particle group; The 2-opt operation is performed on all updated particle groups to calculate the fitness between particles, and update the particle best position and the group best position.

The current group best position is output when the stop conditions for iterative loop are reached, and then the track deflection of the current course data is corrected according to constraint factors.

The optimal route for the USV is finally obtained.

In the specific implementation, in the said route planning controller, the initialized particle swarm parameters include the number of initialized populations, the maximum number of iterations, the fitness function, and the velocity correlation coefficients of particles; the calculation function of fitness is the reciprocal of the route length.

The said constraint factors are converted from the weather and sea wave information of the real-time environment where the USV is located.

The weather and sea wave information of the environment where the USV is located is collected by the ultrasonic weather sensor, including the height, flow velocity and wavelength of the sea wave.

The weather and sea wave information are converted into constraint factors and applied to correct the trajectory of USVs.

The said constraint factors are mainly the sea wave force.

Since the force acting on the vessel in water is mainly the sea wave force, the sea wave force model is used as the constraint factor to correct the track deflection.

The function of constraint factors can be written as: _ 31172 2 Fo =-0.0073V;"hV, +0.0057V , 3 3 Mo == Ach 5 Wherein, h is the height of the sea wave; Vi is the velocity of the sea wave; Ao is the wavelength of the sea wave; Mo is the mass of the sea wave.

In the specific implementation, in the route planning controller, the steps of performing the 2-opt operation on all updated particle groups are shown as follows:

lu101833 Randomly select two non-adjacent positions from each updated particle group. The route segments between those non-adjacent positions are completely reversed and connected back to the original route, resulting in a new route. Repeat this process until the shortest route in each updated particle group is found.

In the specific implementation, the USV model is 1.8 meters long and 0.9 meters wide. Fig. 9 shows the schematic diagram of the structure of route planning system for USVs using the particle swarm optimization based on greedy mechanism. When selecting planning points, the route planning controller generates a feasible trajectory for the USV using the particle swarm optimization based on greedy mechanism. The planned trajectory will be transmitted to the automatic pilot controller together with the navigation information in the bow direction and the USV position data collected by the data acquisition unit composed of multiple sensors such as electronic compass and GPS, and then the course and speed of the vehicle are determined through the closed-loop control until the next target point is arrived.

Tab. 3: Result of simulation of three planning points by ACO and IPSO “Q Algorithm ma; D(m) © ACO 47 1100.23 30 IPSO 27 1100.23 ACO 59 1214.34 40 [PSO 33 1200.35 ACO 78 1726.94 50 IPSO 58 1661.23 The population is generally initialized in a random manner when the traditional particle swarm optimization is applied to the TSP or route planning problem of USV. A large number of infeasible solutions will be generated, thus limiting the computational efficiency. Therefore, this embodiment introduces the greedy mechanism and the 2-opt operation based on a combined strategy. Its purpose is to improve the probability of obtaining the optimal solution and improve the quality of the solution while optimizing the traditional algorithm. In order to verify the effectiveness andlu101833 reliability of the improved algorithm, Monte Carlo simulations are carried out on the eight TSPLIB instances, and application tests are conducted on the self-developed USV.

Embodiment 3, the present embodiment provides a computer-based readable memory medium that stores a computer program. When the program is executed by the processor, the steps in the said route planning method for USVs using the particle swarm optimization based on greedy mechanism described above are realized. Embodiment 4, the present embodiment provides a computer equipment, including a memory, a processor, and a computer program that are stored on the memory and can be executed on the processor. When the said processor executes the said program, the steps in the said route planning method for USVs using the particle swarm optimization based on greedy mechanism described above are realized.

Those skilled in the art should understand that the embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may use the forms of hardware embodiment, the software embodiment, or the combination of software and hardware embodiments. Moreover, the present invention may use the forms of the computer program product implemented on one or more available memory media of computer (including but not limited to disk memory and optical memory) containing available program codes of computer.

The present invention is described with reference to flow charts and/or block diagrams of methods, devices (systems), and computer program products of embodiments of the present invention. It should be understood that the combination of each process and/or block in the flow charts and/or block diagrams can be realized by computer program commands. These computer program commands can be provided for the processor of a all-purpose computer, special-purpose computer, embedded processor, or other programmable data processing equipment to produce a machine that enables the commands executed by the processor of a computer or other programmable data processing equipment to generate a device that can realize the functions specified in one process or multiple processes in a flowchart and/or one block or multiple blocks in a block diagram.

These computer program commands can also be stored in a computer readable memory that is able to guide a computer or other programmable data processing equipment to work in a specific manner, so that the commands stored in the computer readable memory generate an article of

| 22 lu101833 manufacture including command device that can realize the functions specified in one process or multiple processes in a flowchart and/or one block or multiple blocks in a block diagram.

These computer program commands can also be loaded onto a computer or other programmable data processing equipment for a series of operating steps on it to produce computer-implemented processing, so that the commands that are executed on the computer or other programmable equipment provide steps for implementing the functions specified in one process or multiple processes in a flowchart and/or one block or multiple blocks in a block diagram.

Those skilled in the art understand that all or part of the processes for realizing the methods of the foregoing embodiments can be completed by hardware related to computer program commands.

The said program can be stored in a computer readable memory medium.

When the program is executed, the processes of embodiments for the above-mentioned methods may be included.

Wherein, the mentioned storage medium can be magnetic disk, optical disk, Read-Only Memory (ROM) or Random Access Memory (RAM), etc.

The description above is only the preferred embodiments of the present invention but not to limit the present invention.

For the technicians in this field, the present invention can have any alteration and change.

Any modification, equivalent replacement and improvement made within the spirit and principle of the invention shall be included in the scope of protection of the claim of the invention.

Claims

| 1. A route planning method for unmanned surface vehicles(USVs) using particle swarm optimization based on greedy mechanism, the method comprising: obtaining the current position, course data and target position of the USV; calculating the optimal position for the USV to sail from the current position to the target position by the particle swarm optimization based on greedy mechanism, and then modifying the trajectory deflection of the USV's course data according to the preset constraint factor, and finally getting the optimal route of the USV; in particular, the process of calculating the optimal position for the USV to sail from the current position to the target position by the particle swarm optimization based on greedy mechanism: establishing a greedy black box, initialize the particle swarm parameters based on the position of the USV, and generating an initial particle swarm; the initial particle swarm is composed of several initial routes of the USV, and the particle is the location point on the route; entering the iterative loop, and dividing particles in the initial particle swarm of each iteration into preset number of particle groups; screening out the two particles O (1,2) with the lowest fitness in each particle group, and generating two new particles O' (1,2) using the greedy black box; the fitness is the reciprocal of the route length; comparing the fitness of O (1,2) and O' (1,2) to keep the two particles with high fitness and update the particle group; performing the 2-opt operation on all updated particle groups and finding the shortest route in each updated particle group, updating the optimal particle position and the optimal group position until the conditions for stopping the iteration loop are reached, and outputting the current optimal group position.
2. The route planning method for USVs using particle swarm optimization based on greedy mechanism according to claim 1, wherein the initialized particle swarm parameters include theeee ae . . . . . lu1p1883 number of initialized populations, the maximum number of iterations, the fitness function, and the velocity correlation coefficients of particles.
3. The route planning method for USVs using particle swarm optimization based on greedy mechanism according to claim 2, wherein the stop condition of the iteration loop is that the maximum number of iterations is reached, or the fitness of the current group is minimum.
4. The route planning method for USVs using particle swarm optimization based on greedy mechanism according to claim 1, wherein the said constraint factor is the sea wave force, which is a known function of sea wave height and sea wave velocity.
5. The route planning method for USVs using the particle swarm optimization based on greedy mechanism according to claim 1, wherein the steps of performing the 2-opt operation on all updated particle groups include: randomly select two non-adjacent positions from each updated particle group. The route segments between those non-adjacent positions are completely reversed and connected back to the original route, resulting in a new route. Repeat this process until the shortest route in each updated particle group is found.
6. A route planning system for USVs using the particle swarm optimization based on greedy mechanism, wherein: data acquisition module, which is used to acquire the current position, course data and target position of the USV; route planning module, which is used to calculate the optimal position for the USV to sail from the current position to the target position by the particle swarm optimization based on greedy mechanism, and then modify the trajectory deflection of the USV's course data according to the preset constraint factor, and finally get the optimal route of the USV; in particular, the process of calculating the optimal position for the USV to sail from the current position to the target position by the particle swarm optimization based on greedy mechanism: establishing a greedy black box, initialize the particle swarm parameters based on the position of the USV, and generating an initial particle swarm; The initial particle swarm is composed ofseveral initial routes of the USV, and the particle is the location point on the route; lu101833 entering the iterative loop, and dividing particles in the initial particle swarm of each iteration into preset number of particle groups; screening out the two particles O (1,2) with the lowest fitness in each particle group, and generating two new particles O' (1,2) using the greedy black box; The fitness is the reciprocal of the route length; comparing the fitness of O (1,2) and O' (1,2) to keep the two particles with high fitness and update the particle group; performing the 2-opt operation on all updated particle groups and finding the shortest route in each updated particle group, updating the optimal particle position and the optimal group position until the conditions for stopping the iteration loop are reached, and outputting the current optimal group position.
7. The route planning system for USVs using the particle swarm optimization based on greedy mechanism according to claim 6, wherein in the route planning module, the initialized particle swarm parameters include the number of initialized populations, the maximum number of iterations, the fitness function, and the velocity correlation coefficients of particles.
8. The route planning method for USVs using the particle swarm optimization based on greedy mechanism according to claim 6, wherein in the route planning module, the steps of performing the 2-opt operation on all updated particle groups are shown as follows: randomly select two non-adjacent positions from each updated particle group; the route segments between those non-adjacent positions are completely reversed and connected back to the original route, resulting in a new route; repeat this process until the shortest route in each updated particle group is found.
9. A computer readable memory medium on which a computer program is stored, wherein the steps in any route planning method for USVs using the particle swarm optimization based on greedy mechanism according to claims 1-5 are realized when the program is executed by a Processor.
10. A computer device, comprising a memory, a processor, and a computer program stored on the memory and executed on the processor, wherein the steps in any route planning method for

USVs using the particle swarm optimization based on greedy mechanism according to claiméu103833 are realized when the said processor executes the said program.