LU101832B1 - Route Planning Method and System for Unmanned Surface Vehicles Based on Improved Particle Swarm Optimization - Google Patents
Route Planning Method and System for Unmanned Surface Vehicles Based on Improved Particle Swarm Optimization Download PDFInfo
- Publication number
- LU101832B1 LU101832B1 LU101832A LU101832A LU101832B1 LU 101832 B1 LU101832 B1 LU 101832B1 LU 101832 A LU101832 A LU 101832A LU 101832 A LU101832 A LU 101832A LU 101832 B1 LU101832 B1 LU 101832B1
- Authority
- LU
- Luxembourg
- Prior art keywords
- optimization
- particle swarm
- optimal
- iterations
- usv
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/20—Instruments for performing navigational calculations
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/26—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network
- G01C21/34—Route searching; Route guidance
- G01C21/3446—Details of route searching algorithms, e.g. Dijkstra, A*, arc-flags, using precalculated routes
Landscapes
- Engineering & Computer Science (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Automation & Control Theory (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Feedback Control In General (AREA)
Abstract
The present invention discloses a route planning method and system for unmanned surface vehicles based on improved particle swarm optimization. The method includes: acquiring several positions to be passed by the USV; acquiring an optimal traverse route of the USV by the iterative calculation of several positions to be passed by the USV based on the improved particle swarm optimization; wherein, the improved particle swarm optimization is obtained based on the optimization strategies of linearly decreasing inertia weight, adaptive control acceleration coefficient, and random grouping inversion; controlling the USV to move according to the optimal moving route.
Description
Route Planning Method and System for Unmanned Surface Vehicles Based on Improved’! 01832 Particle Swarm Optimization Field of Technology The present disclosure involves the technical field of unmanned surface vehicles(USVs) route planning, especially the route planning method and system for USVs based on an improved particle swarm optimization. Background of the invention The statements in this section only involve the related background information of the present disclosure, and do not necessarily constitute the prior art.
It is well known that the integration of multiple sensors is an important issue for autonomous navigation of unmanned aerial vehicles, especially when unexpected changes are made in the actual environment. With the help of multiple sensors including temperature and humidity sensor, collision sensor, flow sensor, and displacement sensor, unmanned aerial vehicles have been effectively used in sounding detection, environmental monitoring, underwater acoustics, maritime rescue, target tracking, and water monitoring. In these cases, all the perceptual information from multiple sensors is combined and effectively utilized to generate the ideal trajectory of unmanned aerial vehicles, which is often expressed as a traveling salesman problem (TSP).
TSP was proved to be a typical non-deterministic polynomial combinatorial optimization problem in 1979. Its goal is to design a shortest route for travelers to travel every city without repeat and eventually return to the departure city. Since the search space tends to be infinite and complex, traditional accurate algorithms such as enumeration method cannot approach exact solutions within accurate computing time. Therefore, it is required to develop new algorithms with self-organizational and self-adaptive capabilities to find appropriate solutions while giving away the optimality, accuracy and completeness of the operating speed. Inspired by natural evolution models and adaptive population evolution, collective intelligence methods include genetic algorithm, particle swarm optimization (PSO), ant colony optimization (ACO), artificial fish swarm algorithms, and artificial bee colony algorithm. TSP has witnessed a rapid development.
Particle swarm optimization is an evolutionary meta-heuristic technique proposed by Eberhart and Kennedy in 1995. It solves the optimization problem by moving a set of candidate solutions (called particles) at a certain speed in a multi-dimensional search space. Each solution is evaluatedby a fitness function. The movement of all particles is dynamically guided by their own experience 1832 and the experience of the whole group. Finally, the artificial bee colony algorithm is expected to provide the most satisfactory solution. The PSO has the advantages of fast convergence, simple parameter setting, and easy realization, which make it have been widely used in the fields of function optimization, neural network training, and fuzzy system control.
As we all know, the performance of particle swarm optimization largely depends on the appropriate balance between exploration (i.e. searching for a wider space) and development (i.e. moving to a local optimum). The adjustment of PSO parameters has a significant impact on the optimal performance. Thus, many studies have put emphasis on finding appropriate parameters to improve the effectiveness of algorithms.
Furthermore, with the rapid development of intelligent algorithms and autonomous navigation technology, particle swarm optimization has also been applied for vehicle route planning. However, the inventor found that the USV route planning based on the particle swarm optimization in the prior art has the technical problems such as too long planned route, the tendency of getting into local optimum, and insufficiently accurate route planning result. Summary of the Invention In order to solve the shortcomings of the prior art, the present disclosure provides a route planning method and system for USVs based on an improved particle swarm optimization.
First, the present disclosure provides a route planning method for USVs based on an improved particle swarm optimization; the route planning method for USVs based on an improved particle swarm optimization, including: acquiring several positions to be passed by the USV; acquiring an optimal traverse route of the USV by the iterative calculation of several positions to be passed by the USV based on the improved particle swarm optimization; the improved particle swarm optimization is obtained based on the optimization strategies of linearly decreasing inertia weight, adaptive control acceleration coefficient, and random grouping inversion; controlling the USV to move according to the optimal moving route.
Second, the present disclosure provides a route planning system for USVs based on an improved particle swarm optimization;
the route planning system for USVs based on an improved particle swarm optimizatibtf” 832 including: the acquisition module, which is configured as: Acquire several positions to be passed by the USV; the acquisition module of optimal moving route, which is configured as: acquire an optimal traverse route of the USV by the iterative calculation of several positions to be passed by the USV based on the improved particle swarm optimization; wherein, the improved particle swarm optimization is obtained based on the optimization strategies of linearly decreasing inertia weight, adaptive control acceleration coefficient, and random grouping inversion; the control module, which is configured as: control the USV to move according to the optimal moving route.
Third, the present disclosure also provides an electronic device, comprising a memory, a processor, and computer commands stored on the memory and executed on the processor. When the said computer commands are executed by the processor, the method described in the first aspect is finished.
Fourth, the present disclosure also provides a computer readable memory medium for storing computer commands. When the said computer commands are executed by a processor, the method described in the first aspect is finished.
Compared with the prior art, the present disclosure has the following beneficial effects: (1) the route length is effectively shortened and the robustness is enhanced through the iterative adjustment of important parameters such as acceleration coefficient and inertia weight; (2) through the random grouping inversion strategy, the diversity of the swarm is maintained; the convergence rate and global convergence of the algorithm are accelerated; premature populations are avoided, and the accuracy of solutions is maintained; (3) by combining the traditional particle swarm optimization with three optimization strategies, a feasible route with a satisfactory length and without self-crossover is generated. Brief description of the drawings Drawings forming part of the present disclosure are used to provide further understanding herein. The illustrative embodiments herein and their explanations are used to interpret the present disclosure and do not constitute undue restrictions herein. Fig. 1 is a schematic diagram of particle position update in the particle swarm optimization ofthe first embodiment. lu101832 Fig. 2 (a) through Fig. 2 (e) are the solution distribution of each algorithm of the five planning points of the first embodiment.
Fig. 3 (a) through Fig. 3 (e) are the solution distribution of each algorithm of the first embodiment under 5 overall scales.
Fig. 4 (a) through Fig. 4 (e) are the evolution curves of the optimal route distances against the number of iterations for each algorithm of the first embodiment.
Fig. 5 (a) through Fig. 5 (e) are the best routes for the five TSPLIB instances of the first embodiment.
Detailed description of the embodiments It is important to note that the following detailed description is illustrative and aims to provide a further description of the present disclosure. 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 the present disclosure 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 the present disclosure. 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 based on an improved particle swarm optimization; the route planning method for USVs based on an improved particle swarm optimization, including: S1: Acquiring several positions to be passed by the USV; S2: Acquiring an optimal traverse route of the USV by the iterative calculation of several positions to be passed by the USV based on the improved particle swarm optimization; wherein, the improved particle swarm optimization is obtained based on the optimization strategies of linearly decreasing inertia weight, adaptive control acceleration coefficient, and random grouping inversion; S3: Controlling the USV to move according to the optimal moving route.
As one or more embodiments, the specific steps of the improved particle swarm optimization include:
S201: Initializing and selecting the population size and the maximum number of iteratib {1832 defining the fitness function; setting acceleration coefficients cl and c2 based on the optimization strategy of adaptive control acceleration coefficient; 5 setting the inertia weight w based on the optimization strategy of linearly decreasing inertia weight; S202: Evaluating the initial fitness value for the initialization velocity and position each particle, and record the optimal position of initial individual and the optimal position of particle swarm; S203: Judging whether the maximum number of iterations is reached or the error value is less than the set minimum error value; if the maximum number of iterations is reached or the error value is less than the set minimum error value, it will be ended; if the maximum number of iterations is not reached, or the error value is greater than or equal to the set minimum error value, then proceed to S204; i S204: Updating individual velocity and individual position for each particle, and evaluating new fitness values; dividing the population into several sub-populations based on the optimization strategy of random grouping inversion, and independently evolving each sub-population; updating the optimal position of each particle and the optimal position of the particle swarm, and returning to S203.
As one or more embodiments, the acceleration coefficients c, and czareset based on the optimization strategy of adaptive control acceleration coefficient.Specific steps include: K = P/N (6) C1 = (Cimax — Camin)K + Cimin (7) Ca = (C2max — C2min)K + C2min (8) Wherein, P represents the number of successfully converged particles in the primary iteration; K refers to the evaluation coefficient; K is the ratio of the number of successfully converged particles to the initial swarm size in the primary iteration; N represents the total number of particles; cl and c2 are acceleration coefficients; Cimax 1S the maximum value of C1; Ciminis the minimum value of Ci; €zmax is the Maximum value of C2; Czminis the minimum value of ca.
As one or more embodiments, set the inertia weight w based on the optimization strategy forlinearly decreasing inertia weight. Specific steps include: u101832 dynamically adjusting the inertial weight w in the form of linear decrease in the iteration process: W = Wmax 7 MX (Wmax 7 Wmin)/M (3) Wherein, Wmax is the maximum value of the inertial weight W; Win is the minimum value of the inertial weight w; m is the current number of iterations, and M is the maximum number of iterations.
As one or more embodiments, divide a single population into several sub-populations based on the optimization strategy of random grouping inversion; Specific steps include: disrupting the order of all the particles, and independently evolving them with four particles in one group, and obtaining a new particle swarm after all particles are evolved. As one or more embodiments, update the optimal position of each particle and the optimal position of the particle swarm. Specific steps include: vit = wolf + rl (PIS — X) + crt (Bly — xis) (1) xP = x + pt (2) wherein, m and s represent the current number of iterations and the s-th dimension, respectively; rl and r2 are random numbers uniformly distributed between 0 and 1, and w is the control parameter of PSO. In order to overcome the shortcomings of traditional particle swarm optimization such as premature and parameters that are determined based on human experience, improve the accuracy of the solution and the robustness of the algorithm, the present disclosure proposes three optimization strategies: Linearly decreasing inertia weight method, adaptive control acceleration coefficient, random grouping inversion. The Monte Carlo simulation method is used to carry out a simulation study of 5 TSPLIB instances, and the application test of autonomous USV navigation system based on multi-sensor data is carried out. The results show that the acceleration coefficient of adaptive control plays a vital role in reducing the route length, and the linearly decreasing inertia weight is conducive to improving the robustness of the algorithm. Besides, the random grouping inversion optimizes the local search capabilities. By dividing a single population randomly into several sub-populations, the population diversity is maintained. The combination of particle swarm optimization with these three strategies can show the best performance with the shortest trajectoryand stronger robustness, although it takes more time to maintain the accuracy of solutions and avoid 832 getting into the local optimum. Experimental results prove the validity and effectiveness of such method.
To design a multi-target optimization route for autonomous underwater vehicles in a dynamic environment, the particle swarm optimization is used to find a suitable temporary waypoint and an optimal route is generated combined with the waypoint guidance. To avoid premature populations and route self-crossover, and enhance the robustness, the present disclosure proposes three improved algorithms based on the PSO algorithm, in combination with optimization strategies such as linearly decreasing inertial weights, adaptive control acceleration coefficients, and random grouping inversion. First, 100 times of Monte Carlo simulations are carried out on 5 TSPLIB instances, and the effectiveness of each improved algorithm in terms of route length, calculation efficiency, and algorithm robustness is compared. The improved particle swarm optimization is applied to the independently developed USV navigation, guidance and control system based on multi-sensor data.
The traditional particle swarm optimization is a stochastic optimization method based on population. At the beginning of the evolution procedure, the PSO algorithm randomly generates N candidate solutions (i.e., N particles) in the S-dimensional search space. The position and velocity of the i-th particle can be represented by vectors Xi= (xi1, Xi, … Xis)" and Vi = (vi, Vi, ..., Vis)", respectively. The fitness function is defined as 1/D (D is the route length), in terms of the TSP and route planning problem. For each time of iteration, all particles guide their movement depending on two kinds of experiences: One is the best position (Pis) known so far for the individual, and the other is the best position (Pgs) known so far for the whole particle swarm. Accordingly, the velocity and position of each particle are updated in accordance with formulas (1) and (2).
vig tt = wri + CTP — Xis) + card (Pas — x5) (1) x = x0 + pL (2) Wherein, m and s represent the current iterations and s-dimension, respectively. rl and r2 refer to the uniform random numbers between 0 and 1. cl, c2 and w refer to the controls parameter of PSO, which are known as personal cognitive coefficient, social cognition coefficient and inertia weight, respectively.
Care must then be taken that there are three speed terms on the right of the formula (1). wv;}It
| 8 . . . . . . . . . . lu101832 is the inertial component that moves particles in the original direction of the previous iteration.
Inertia weight w, which may make a difference in the capability of global search and algorithm convergence, and is usually set at 0.8 - 1.2. The er" (PX — xy), referred to as personal cognition, permits the particles to move in accordance with the memory of its optimal position.
The cr" (Pe — xi3), referred to as social cognition, guides particles to move towards the optimal position in the population on the basis of communication with other particles.
The acceleration coefficients c; and c play a vital role in balancing the impact of individual cognition and social cognition on guiding particles towards the optimal solution.
The values for cl and c2 are recommended to be 2. Besides, there were reports that the random characteristics of rl and r2 may diminish the influences of most well-known position of individuals and groups on velocity update.
And maintain population diversity as well as prevent precocious population to a certain extent.
What is shown in Fig. 1 is the changes of particle position in two successive iterations.
In the case of maximum iteration(M) or minimum error threshold, the algorithm process is stopped.
The pseudo code for particle swarm optimization is as shown in Tab. 1. Tab. 1 Pseudo Codes of Particle Swarm Optimization Algorithm: Traditional Particle Swarm Optimization for TSP Problem Population Size and Maximum Iterations are chosen Define fitness function Preset acceleration coefficient (cl, c2), and inertial weight (w) For each particle Initialize velocity and position Evaluation of initial fitness value Records of initial P;s and Pgs end while fails to meet the maximum iteration or minimum error criterion For each particle New velocity is calculated with formula (1) New positions are updated with formula (2) Evaluation of new fitness value Updated P;s and Pend end end Inertial weight of linear decrease: Inertial weight w reflects the influence of historical velocities on the current velocity of eachparticle and the capability of balancing the current and global search. 4101832 When w=0, in accordance with formula (1), the current cognition of particle best position(Pis) and global best position (Pgs) determines the particle velocity.
A particle may remain stationary if it is in its current Pg, while other particles may fly at the weighted velocity of Pis and Pgs. | In this case, the entire population may be pulled towards the current Pgs and converged to the local optimum.
On the contrary, all particles tend to explore more spaces by means of inertial component. Thereby, in the case of optimization problems such as function optimization, neural network training and fuzzy system control, etc., w value shall be adjusted to balance the algorithm ability of local search and global search.
Furthermore, the global search behavior, especially convergence behavior, may be affected by the inertia weight w.
By and large, the smaller value of w makes for speeding up the convergence of the globally optimal solution, and larger value of w makes for exploring the whole search space.
For better global search capability in the early iteration process, improving the local utilization in the later iteration process to prevent being trapped into local optimum, the inertia weight w is dynamically adjusted in the form of a linear decrease in the iteration process, in accordance with the formula (3).
W = Wmax — MX (Wmax — Wmin)/M (3) Wherein, Wmax and Win represent the maximum and minimum of inertia weight w, respectively.
Adaptive control acceleration coefficient: The acceleration coefficients cl and c2 reflect the information exchange among particles, which determine the distance that the particle moves towards the target solution in one iteration under the guidance of personal cognition and social cognition.
The small value of acceleration coefficients keeps the particles away from the target region, while the large values makes the particles move rapidly towards the target region, but eventually off the target region.
When cl and c2 equal to 0, the particles may fly at its current velocity until it reaches theborders of the search space. Therefore, there are more difficulties in finding a satisfactory solution! 882 in a limited search space.
If cl equals to 0, the particles may lose the cognitive function. Given that interactions among ; particles can enlarge the search space, but it is easy to be trapped into local optimum in terms of the | 5 complicated optimization problems.
When c2 equals to 0, there are no information exchanges in the swarm. Each particle works independently. It is almost impossible to give an optimal solution.
The values of cl and c2 remain unchanged throughout the evolution of traditional PSO. Nonetheless, the fixed setting features its inherent limitations: Larger values may cause quick 10 convergence of each particle to local optimal, while smaller values may make each particle away from the target area.
The concept of iterative linear variation velocity coefficients is used therefore, and larger cl and smaller c2 relatively are used in the early phase of the iteration. With the increment of iterations, the linearity of cl value is decreased, while of c2 value is increased, as shown in the formula (4) and (5).
Research shows that the acceleration coefficient of linear change makes for reducing the premature convergence probability in the early iterations, and improving the convergence performance at the end of iteration.
C1 = (Cumin — Cimax) 77 + Cimax (4) C2 = (Camax — Comin) 77 + Comin (5) Wherein, max and min subscripted herein refers to the maximum and minimum values of cl and c2.
However, in the whole evolution process, the acceleration coefficients have limited the influences on the convergence of the algorithm when its linearity changes. For instance, the social information poses a great influence on the search efficiency later period of algorithm, while it cannot be realized by the simple linear change of acceleration coefficient in time.
Therefore, the present disclosure may introduce an evaluation parameter K based on the convergence degree of the algorithm, of which the value is defined as the ratio of the number of successfully converging particles (known as favorable particles) to the size of initial swarm in one iteration, as shown in Equation (6). Then according to equation (7) and equation (8), the evaluationparameter (k) is used to control adaptively the change rate of acceleration coefficient. 1101832 K=P/N (6) ¢1 = (Cimax — Camin)K + Camin (7) Ca = (C2max — Camin)K + C2min (8) Wherein, P represents the number of successfully converged particles in the primary iteration.
The evaluation parameter K is used in the strategy of adaptive control acceleration coefficient to correlate the value of the acceleration coefficient with the optimization state. With the increment of iterations, the favorable particles in the swarm are also increased, which may amplify its affection on the entire swarm. Thereby, the excellent solutions shall be given more protections as possible, to prevent local optimal.
Random Group Inversion: The CPSO is evolved with a population of all particles, which may cause the phenomenon that all particles clump together in one position, and stop exploring other areas of the search space therefore. For preventing premature population, the concept of random group inversion is proposed, which is added in each iteration before updating Pis and Pg. The single population is divided into several sub-populations, and evolved independently, which enhances the diversity of population and accelerates the global convergence of the whole population.
As for the number of particles in the sub-population, the preliminary study shows that larger number may lower the internal capability of value-based selection, while the smaller number may weaken the role of grouping mechanism. Eventually, set to 4. In other words, four particles may be sorted at random to generate a sub-population.
On basis of the random group strategy, an algorithm that simulates the inversion operation of biological evolution is proposed. Evaluation of the adaptability of each particle shows that all four particles clump together around the local optimal value found in each sub-population, and then inversion is made for the particle with the minimum path value among the four particles, to generate a new particle and replace any two original particles in the sub-population, wherein the TSP travel sequences of the two randomly selected inversion points are exchanged.
The local optimal value refers to the optimal value under the circumstances by calculating the path values of four particles and giving the minimum path value.
The inversion is conducted for the particle with minimum path value among the four partidfdd) 832 in the way that two inversion points are determined by random numbers and the TSP sequence of two inversion points is changed to generate a new particle and replace any two of the original to complete.
The said inversion point refers to: TSP travel sequence represented by the random number.
As a matter of fact, the Darwin's theory of evolution serves as a basis for the strategy of random group inversion: Internal competition and uncertain mutation of the population. Theoretically, the internal competition of population is a process of optimum selecting, that is, survival of the fittest. Inversion refers to an uncertain mutation, which makes for maintaining the population diversity. In the end, there are not only the most suitable individuals in each sub-population, but also inversion-based variants retained in the particle swarm pools. Therefore, the strategy can enhance the population diversity and improve the effectiveness of population optimization. The pseudo-code for random group inversion is shown in Tab. 2.
Tab. 2 Pseudo Code for Random Group Inversion Algorithm: Random allocation for each sub-population Finding out the fittest particle in the four particles Randomly selecting the inversion points Inversion Update Pg; end Three algorithms of improvement strategy are proposed on the basis of the particle swarm optimization: Adaptive control acceleration coefficient algorithm (APSO), adaptive control acceleration coefficient algorithm and linear decrease inertia weight algorithm (AWPSO), which integrates the merits of adaptive control of acceleration coefficient, the linear decrease of inertia weight and random group inversion (AWIPSO). For the purpose of reducing the randomness of particle swarm algorithm in MATLAB operating environment, the Monte Carlo simulation method is used in the present section to compare the performance of various algorithms to TSP from three aspects including planning points, swarm size and computational efficiency. To prevent the influence of the computer model on the algorithm performance, all simulations are done on the same computer with 8.0gb memory (Intel(R)Core(TM)17-7700HQ CPU@2.80GHz). Moreover, all instances are from TSPLIB.
. . . . lu101832 Comparative Study of Different Planning Points: Five instances taken into consideration in TSPLIB refer to Burmal4, ulysses22, eil51, eil76, and rat99. Accordingly, the number of planning points (Q) is 14, 22, 51, 76 and 99, respectively, and the maximum iteration (M) is set to 100, 200, 1,600, 2,000 and 2,000, respectively. The population size (R) of each algorithm is set to 500. As to CPSO, the acceleration coefficients cl and c2 are at a value of 2 and w is at a value of 0.9. For the improved algorithm, the variation range of individual cognition coefficient cl is set as 0.9-1.2 and of social cognition coefficient c2 as 0.2-1.0 during the evolution. For the linear decrease inertia weight w, the value changes from 0.9 to 0.4. Reference is made to Tab. 3 for the parameter settings of each algorithm.
Tab. 3 Parameters Settings of Each Algorithm Algorithm Parameters Settings Individual cognitive coefficient c; 2 CPSO Social cognitive coefficient c2 2 Inertia weight w 0.9 Individual cognitive coefficient cı Adaptive control according to formula (7), 0.9-1.2 APSO Social cognitive coefficient c2 Adaptive control according to formula (8), 0.2-1.0 Inertia weight w 0.9 Individual cognitive coefficient c; Adaptive control according to formula (7), 0.9-1.2 AWPSO Social cognitive coefficient cz Adaptive control according to formula (8), 0.2-1.0 Inertia weight w Linear decrease according to formula (3), 0.9-0.4 Individual cognitive coefficient cı Adaptive control according to formula (7), 0.9-1.2 AWIPSO Social cognitive coefficient c; Adaptive control according to formula (8), 0.2-1.0 Inertia weight w Linear decrease according to formula (3), 0.9-0.4 Monte Carlo simulations are done 100 times for each TSP instance, to obtain the optimal path distance (D) data set of four algorithms. The comparison results are shown in block diagram and whisker plot (as shown in Fig. 2 (a) - (e)). Legends and explanations regarding box and beards refer to the works of MLE. Spear. In each block diagram, a range bar shows the quartile range of data set, which to a certain extent indicates the degree of data dispersion and robustness of the algorithm. The intermediate values and mean values are marked in a bar chart with line and plus sign, meanwhile the beard is extended to both sides of the bar, with the two ends showing the optimal value and the worst value, respectively.
Furthermore, there are data sets of each algorithm and detailed statistics of each planning point listed in Tab. 4, including the known TSPLIB optimal solution, the mean value of the worst, best and optimal path distances. In addition, the standard difference is computed to present the distance of data set to the mean value, which shows the robustness of the algorithm.
In the burmal4 example in Fig. 2(a), it shows that the proposed optimization strategy CAR) 832 reduce the mean value of D, and allow the data set closely clump together. However, in terms of path distance and algorithm robustness, the three improved algorithms show less than 0.4% difference.
When planning points (Q) are increased (referring to Fig. 2(b)-(e)), the adaptive control acceleration coefficient, linear decrease inertia weight and random group inversion gradually show their merits. For Q=76, the acceleration coefficient strategy of adaptive control makes a real difference in reducing path length when comparison is made between APSO and CPSO. Meanwhile, the inertia weight of linear decrease poses main influences on the dispersion of data. The standard error of AWPSO is 52.48m, 10.7% lower than APSO. In comparison, the average shortest path distance of AWIPSO is 590.1m, and its standard error is 12.4m, with obvious superiority, which shows that the random group inversion strategy is of great significance in reducing path length and improving the robustness of the algorithm. Besides, when Q=99, the AWIPSO has the optimal value of 1,248m, only 3% greater than the known optimal solution of TSPLIB.
A Comparative Study of Different Population Sizes; The TSP instance of eil51 planning points are used as the working condition. Five population sizes (R) are 300, 400, 500, 600 and 700, respectively. And the maximum iterations (m) of each algorithm are set as 500. Other parameters of each algorithm, such as cl, c2 and w, are identical to those in the comparative study of different planning points.
Monte Carlo simulations are made 100 times for each algorithm and each population. The comparison results are shown in the five block diagrams and whisker plots in Fig. 3 (a) - (e). There are detailed statistics of the optimal path distance listed in Tab. 5, including the summary of four datasets.
When the population size is 300, as shown in Fig. 3 (a), the respective advantages of three optimization strategies can be similarly summarized as shown in Fig. 2 (a) - (e). The acceleration coefficient of adaptive control makes for reducing the path distance. In comparison of APSO and AWPSO, the linear decrease inertia weight poses a certain influence on 2.1% decrease of data dispersion. Of the random group inversion, the average optimal path length is 457.0 m, and the standard error is 9.9 m, which shows obvious effects in reducing the path distance and improving the robustness of the algorithm, compared with AWPSO. Moreover, the mid-value is almost
. Ce . 9" 832 identical to the mean value in each bar, which implies that all algorithms can generate unifo | distributed data under eilS1. Tab. 4 Statistical Results of 100-times Optimal Path Distance at 5 Planning Points
LE TT Known optimal . Worst solution Optimal Average value Standard Q solution Algorithm m solution (m m deviation (m CPSO 36.61 30.88 31.90 1.00
30.88 APSO 32.21 30.88 31.14 0.36 14 0.88 m AWPSO 32.42 30.88 31.16 0.38 AWIPSO 32.15 30.88 31.03 0.29 CPSO 99.78 76.49 85.45 6.22 74 APSO 80.22 75.50 77.26 1.13 22 m AWPSO 80.05 75.30 77.11 1.10 AWIPSO 77.10 75.30 75.97 0.46 CPSO 935.06 684.73 825.02 45.67 51 426 APSO 759.91 594.27 665.87 38.96 m AWPSO 775.86 597.08 668.68 38.72 AWIPSO 480.14 436.06 455.91 9.37 CPSO 1459.78 1156.23 1315.03 62.90 76 538 APSO 1247.22 948.52 1082.12 58.75 m AWPSO 1231.88 966.33 1071.70 52.48 AWIPSO 625.39 564.86 590.11 12.40 CPSO 4555.13 3602.25 4136.63 194.64 99 1211 APSO 3772.27 2912.69 3359.80 178.09 m AWPSO 3819.77 2998.28 3379.08 173.14 AWIPSO 1394.07 1248.12 1332.97 27.62 Tab. 5. Statistical Results of 100-times Optimal Path Distance under Five Population Sizes R Algorithm Worst solution (m) Optimal solution (m) Average value (m) Standard deviation (m) CPSO 974.27 776.77 872.24 43.50 300 APSO 781.48 601.64 696.70 40.75 AWPSO 829.51 583.85 700.67 39.90 AWIPSO 487.43 434.33 457.03 9.85 CPSO 992.82 699.91 849.28 51.74 400 APSO 770.40 566.11 682.41 37.41 AWPSO 792.31 568.42 676.16 41.31 AWIPSO 495.28 434.84 456.84 10.69 CPSO 930.74 719.46 825.04 40.90 500 APSO 765.91 582.15 663.66 35.23 AWPSO 755.14 584.22 666.19 37.57 AWIPSO 485.47 434.60 457.57 9.75 CPSO 933.20 690.43 793.78 49.21 600 APSO 744.87 557.78 655.20 32.53 AWPSO 746.03 568.09 659.74 38.71 AWIPSO 1480.44 434.9 456.19 9.11 CPSO 941.57 684.87 784.95 42.51 200 APSO 728.26 564.87 639.80 34.03 AWPSO 755.14 559.05 650.16 37.26 AWIPSO 489.62 435.14 456.05 9.23 In a general way, the population size poses influences on the algorithm performance that theglobal optimal distance of each algorithm can be further reduced with the increase of the populaffdy’ 832 size. For instance, in terms of R=700, the average D value of CPSO is 785.0 m, 10% less than that for R = 300. For AWIPSO, when R = 700, the average D is 456.1 m, only 0.2% less than that for R = 300. The results show that the population size has no distinct impact on the performance of algorithm. And then, although each algorithm may change slightly as the population size changes, there are no regular change trends. By contrast, AWIPSO is always the most advantageous algorithm. There are almost no effects of population size on the optimal path length and algorithm robustness. When R=700, the average distance of AWIPSO is 456.1 m and the standard error is 9.2 m, 42.0% and 78.3% less than CPSO, respectively.
Comparison Results of Calculation Efficiency: The evolving curves of TSPLIB instances with five different planning points are given to compare the calculation efficiency of each algorithm. Two main efficiency evaluation indexes are used: Time consumption and convergence rate. The former refers to the time consumption to complete the maximum iterations, while the latter refers to the number of iterations (MCRI) at which the solution converges to the optimal value. It is the evolutionary history of the optimal path distance (D) of each algorithm with respect to the number of iterations (m) in 500 iterations shown in Fig. 4 (a) - Fig. 4 (e). While, all detailed information about the calculation efficiency of each algorithm is listed in Tab. 6.
Generally, in the initial phase, as the iteration increases, the evolving curve of optimal path distance of each algorithm decreases sharply, then shows a slow decrease trend, and finally reaches the critical number level (Mec). The number of planning points and the algorithm used to determine the development of the evolution curve and the global optimal distance. With the increase of planning points (Q), there is increasing tendency shown in the critical number and optimal distance of each algorithm.
Compared with the other algorithms of the five planning points considered in the present section, AWIPSO is converged to the minimum path distance. What's more, when P is less than 22, there is no significant difference between the critical value of AWIPSO and other critical values. But, as Q increases, AWIPSO shall be converged with the most iterations. For example, when Q=76, the moi of AWIPSO is 378, which is 6, 1.9 and 1.4 times that of CPSO, APSO and AWPSO, respectively. The reasons may be that the conditions with more planning points may enhance the
. . ; ug 832 complexity of the path, and are of greater demand for algorithm performance, especially | preventing premature convergence. In other words, during the evolution, the CPSO, APSO and AWPSO may be trapped into local optimization, causing relatively smaller mer, while AWIPSO ensures the accuracy of solutions, and requires more iterations before convergence. In terms of calculation time, when Q=99, AWIPSO takes 6.7, 2.4 and 2.4 times more time than CPSO, APSO and AWPSO to complete the same iterations, respectively. It is necessary to extend the calculation time overhead to a certain extent, for reducing the path distance. Moreover, the optimal paths of five TSPLIB instances using AWIPSO (burmal4, ulysses22, eil51, eil76, rat99) are shown in Fig. S (a) - Fig. 5 (e). The x-coordinate and y-coordinate represent the values of latitude and longitude, respectively. The origin is enclosed in the rectangle, and the arrow shows the direction of the planning path. It is observed that the shape and distance of the path become more complicated as the Q increases. Tab. 6 Simulation Results of Computational Efficiency of Each Algorithm | |main| | Time Com aso |, | | 7 so | es 100 4 amso |, | 5) so |, | 04 2 so, | os 200 2
AWPSO aso |, | 0s aso || us 7 so | 53 1 1600 so | | aa | aso || 17 | 64lu101832 CPSO | 3.0 | wo || ow 02 2000 AWIPSO 23.6 78 ao |] so 7 so |, | wo 10 2000 AWPSO 14.1 22 AWIPSO 33.7 75 Multi-sensor fusion relates to an important problem of USV autonomous navigation. The present disclosure, on the basis of particle swarm optimization, has proposed three optimization strategies to realize USV real-time autonomous navigation in the actual marine environment: Linearly decreasing inertia weight method, adaptive control acceleration coefficient and random grouping inversion. Monte Carlo simulation has been done for five TSPLIB instances, and the application test has been conducted on unmanned ground vehicles (UGV), to reveal their respective or combined advantages. The results are as follows: (1) The influences of favorable particles on the swarm are used in the acceleration coefficient of adaptive control, to improve the capability of early iterative global search and late local search.
This strategy has played a crucial role in reducing path length.
(2) The inertia weight of linear decrease makes for improving the robustness of the algorithm.
(3) The random group inversion optimizes the local search capability, and maintains the population diversity, prevents the premature population, and retains the accuracy of the solution.
(4) Particle Swarm Optimization (PSO) combining these three strategies shows the superiority in generating paths with most satisfactory length and no adaptive crossover. However, more time consumption is required before global convergence.
(5) The more the planning points are, the more complex the generated paths are, and the higher requirements for the algorithm performance are. Whereas, the population size of each algorithm poses irregular influences on path planning, which can be negligible to a certain extent.
Embodiment 2, there is also the path planning system for unmanned ships based on theimproved particle swarm optimization provided in the present embodiment. u101832 The route planning system for USVs based on an improved particle swarm optimization, including: the acquisition module, which is configured as: acquiring several positions to be passed by the USV; the acquisition module of optimal moving route, which is configured as Acquire an optimal traverse route of the USV by the iterative calculation of several positions to be passed by the USV based on the improved particle swarm optimization; wherein, the improved particle swarm optimization is obtained based on the optimization strategies of linearly decreasing inertia weight, adaptive control acceleration coefficient, and random grouping inversion; the control module, which is configured as: controlling the USV to move according to the optimal moving route.
As one or more embodiment(s), the optimal mobile path acquisition module includes: the initialized unit, which is configured as: initializing and select the population size and the maximum number of iterations; defining fitness function; setting acceleration coefficients cl and c2 based on the optimization strategy of adaptive control acceleration coefficient; setting the inertia weight w based on the optimization strategy of linearly decreasing inertia weight; assessment unit 1, which is configured as: evaluating the initial fitness value for the initialization velocity and position each particle, and recording the optimal position of initial individual and the optimal position of particle swarm; decision unit, which is configured as: judging whether the maximum number of iterations is reached or the error value is less than the set minimum error value; If the maximum number of iterations is reached or the error value is less than the set minimum error value, it will be ended; If the maximum number of iterations is not reached, or the error value is greater than or equal to the set minimum error value, then proceed to the second assessment unit; assessment unit 2, which is configured as: updating individual velocity and individual position for each particle, and evaluating new fitness values; dividing the population into several sub-populations based on the optimization strategy ofrandom grouping inversion, and independently evolving each sub-population; lu101832 updating the optimal position of each particle and the optimal position of the particle swarm, and returning to decision unit.
Embodiment 3, the present embodiment also provides an electronic device, comprising a memory, a processor, and computer commands stored on the memory and executed on the processor.
When the said computer commands are executed by the processor, the method described in the first | aspect is finished.
Embodiment 4, the present embodiment also provides a computer readable memory medium for storing computer commands.
When the said computer commands are executed by a processor, the method described in the first aspect is finished.
The description above is only the preferred embodiments herein but not to limit the present disclosure.
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 present disclosure shall be included in the scope of protection of the claim of the present disclosure.
Claims (10)
1. A route planning method for unmanned surface vehicles (USVs) based on improved particle swarm optimization, the method comprising: acquiring several positions to be passed by the USV; acquiring an optimal traverse route of the USV by the iterative calculation of several positions to be passed by the USV based on the improved particle swarm optimization; wherein, the improved particle swarm optimization is obtained based on the optimization strategies of linearly decreasing inertia weight, adaptive control acceleration coefficient, and random grouping inversion; controlling the USV to move according to the optimal moving route.
2. The method according to claim 1, wherein the specific steps of the improved particle swarm optimization include: S201: initializing and selecting the population size and the maximum number of iterations; defining the fitness function; setting acceleration coefficients cl and c2 based on the optimization strategy of adaptive control acceleration coefficient; setting the inertia weight w based on the optimization strategy of linearly decreasing inertia weight; S202: evaluating the initial fitness value for the initialization velocity and position each particle, and recording the optimal position of initial individual and the optimal position of particle swarm; S203: judging whether the maximum number of iterations is reached or the error value is less than the set minimum error value; if the maximum number of iterations is reached or the error value is less than the set minimum error value, it will be ended; if the maximum number of iterations is not reached, or the error value is greater than or equal to the set minimum error value, then proceed to S204; S204: updating individual velocity and individual position for each particle, and evaluating new fitness values; dividing the population into several sub-populations based on the optimization strategy of random grouping inversion, and independently evolving each sub-population; updating the optimal position of each particle and the optimal position of the particle swarm,
and returning to S203. lu101832
3. The method according to claim 2, wherein the acceleration coefficients c,and cz are set based on the adaptive control acceleration coefficient optimization strategy; specific steps include: K = P/N (6) €, = (Cimax 7 Camin)K + C1min (7) Ca = (Camax — C2min)K + Camin (8) wherein, P represents the number of successfully converged particles in the primary iteration; K refers to the evaluation coefficient; K is the ratio of the number of successfully converged particles to the initial swarm size in the primary iteration; N represents the total number of particles; cl and c2 are acceleration coefficients; C;max iS the maximum value of ¢;; Cjmin refers to the minimum of C,; Czmax!éfers to the maximum of C2; Czmin 1S the minimum value of cz.
4. The method according to claim 2, wherein the inertia weight w is set based on the linear decrease inertia weight optimization strategy; specific steps include: dynamically adjusting the inertial weight w in the form of linear decrease in the iteration process: W = Wmax 7 MX Wimax — Wmin)/M (3) wherein, Wmax is the maximum value of the inertial weight w ; Wmin is the minimum value of the inertial weightw; m is the current number of iterations, and M is the maximum number of iterations.
5. The method according to claim 2, wherein a single population is divided into several sub-populations based on a random group inversion optimization strategy; specific steps include: disrupting the order of all the particles, and independently evolving them with four particles in one group, and obtaining a new particle swarm after all particles are evolved.
6. The method according to claim 2, wherein updating the individual optimal position of each particle and the optimal position of the particle swarm; specific steps include: viet = woul + Gr (PF — Xi) + cor (Pe — x) (1) Xi = Xi + vst (2) wherein, m and s represent the current number of iterations and the s-th dimension, respectively; rl and r2 are random numbers uniformly distributed between 0 and 1, and w is the control parameter of PSO.
7. A route planning system for USVs based on the improved particle swarm optimization, thesystem comprising: 101832 a acquisition module, which is configured as: acquiring several positions to be passed by the USV; a acquisition module of optimal moving route, which is configured as acquiring an optimal traverse route of the USV by the iterative calculation of several positions to be passed by the USV based on the improved particle swarm optimization; wherein, the improved particle swarm optimization is obtained based on the optimization strategies of linearly decreasing inertia weight, adaptive control acceleration coefficient, and random grouping inversion; the control module, which is configured as: controlling the USV to move according to the optimal moving route.
8. The system according to claim 7, wherein the optimal mobile path acquisition module includes: an initialized unit, which is configured as: initializing and selecting the population size and the maximum number of iterations; defining the fitness function; setting acceleration coefficients cl and c2 based on the optimization strategy of adaptive control acceleration coefficient; setting the inertia weight w based on the optimization strategy of linearly decreasing inertia weight; assessment unit 1, which is configured as: evaluating the initial fitness value for the initialization velocity and position each particle, and recording the optimal position of initial individual and the optimal position of particle swarm; a decision unit, which is configured as: judging whether the maximum number of iterations is reached or the error value is less than the set minimum error value; if the maximum number of iterations is reached or the error value is less than the set minimum error value, it will be ended; if the maximum number of iterations is not reached, or the error value is greater than or equal to the set minimum error value, then proceed to the second assessment unit; assessment unit 2, which is configured as: updating individual velocity and individual position for each particle, and evaluating new fitness values; dividing the population into several sub-populations based on the optimization strategy of random grouping inversion, and independently evolving each sub-population;
updating the optimal position of each particle and the optimal position of the particle sway 832 and returning to decision unit.
9. An electronic device, wherein it comprises a memory, a processor and a computer instruction stored in memory and running on the processor, which is executed by the processor, while any of the methods defined in claims 1-6 shall be completed.
10. A computer readable storage media, wherein, used to store the computer instructions that are executed by the processor, while any of the methods defined in claims 1-6 shall be completed.
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010170289.7A CN111174794A (en) | 2020-03-12 | 2020-03-12 | Unmanned ship path planning method and system based on improved particle swarm algorithm |
Publications (1)
Publication Number | Publication Date |
---|---|
LU101832B1 true LU101832B1 (en) | 2020-10-05 |
Family
ID=70648534
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
LU101832A LU101832B1 (en) | 2020-03-12 | 2020-06-05 | Route Planning Method and System for Unmanned Surface Vehicles Based on Improved Particle Swarm Optimization |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN111174794A (en) |
LU (1) | LU101832B1 (en) |
Families Citing this family (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113124873B (en) * | 2021-04-09 | 2022-08-16 | 青岛哈船海智科技有限公司 | UUV multi-index constraint three-dimensional route planning method based on marine environment information |
CN114354870A (en) * | 2021-12-30 | 2022-04-15 | 江苏大学 | Unmanned ship shortest path water quality monitoring system and method based on improved particle swarm optimization algorithm |
-
2020
- 2020-03-12 CN CN202010170289.7A patent/CN111174794A/en not_active Withdrawn
- 2020-06-05 LU LU101832A patent/LU101832B1/en active IP Right Grant
Also Published As
Publication number | Publication date |
---|---|
CN111174794A (en) | 2020-05-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111694365B (en) | Unmanned ship formation path tracking method based on deep reinforcement learning | |
LU101833B1 (en) | Route Planning Method and System for Unmanned Surface Vehicles Using Particle Swarm Optimization Based on Greedy Mechanism | |
CN110887484B (en) | Mobile robot path planning method based on improved genetic algorithm and storage medium | |
CN111722643B (en) | Unmanned aerial vehicle cluster dynamic task allocation method imitating wolf colony cooperative hunting mechanism | |
Hao et al. | Dynamic path planning of a three-dimensional underwater AUV based on an adaptive genetic algorithm | |
LU101832B1 (en) | Route Planning Method and System for Unmanned Surface Vehicles Based on Improved Particle Swarm Optimization | |
CN109933067A (en) | A kind of unmanned boat collision prevention method based on genetic algorithm and particle swarm algorithm | |
CN110095120A (en) | Biology of the Autonomous Underwater aircraft under ocean circulation inspires Self-organizing Maps paths planning method | |
CN110926477A (en) | Unmanned aerial vehicle route planning and obstacle avoidance method | |
CN110906935A (en) | Unmanned ship path planning method | |
Wang et al. | 3M-RL: Multi-resolution, multi-agent, mean-field reinforcement learning for autonomous UAV routing | |
Huang et al. | Reinforcement learning for mobile robot obstacle avoidance under dynamic environments | |
Wang et al. | Monitoring trajectory optimization for unmanned surface vessel in sailboat race | |
CN113391633A (en) | Urban environment-oriented mobile robot fusion path planning method | |
Hu et al. | Super eagle optimization algorithm based three-dimensional ball security corridor planning method for fixed-wing UAVs | |
Hill et al. | Autonomous situational awareness for UAS swarms | |
CN113419524A (en) | Robot path learning and obstacle avoidance system and method combining deep Q learning | |
CN110779526B (en) | Path planning method, device and storage medium | |
CN111427368A (en) | Improved multi-target collision-prevention driving method for unmanned intelligent vehicle | |
CN110941263A (en) | Autonomous underwater vehicle path planning method and device | |
CN115774455A (en) | Distributed unmanned cluster trajectory planning method for avoiding deadlock in complex obstacle environment | |
Xin et al. | Improved genetic algorithms based on data-driven operators for path planning of unmanned surface vehicle | |
Keong et al. | Reinforcement learning for autonomous aircraft avoidance | |
Yu et al. | An intelligent robot motion planning method and application via lppo in unknown environment | |
CN118735095A (en) | Harris eagle optimization algorithm and system for heterogeneous carrier cluster route planning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
FG | Patent granted |
Effective date: 20201005 |