A Genetic Algorithm Continuous Search Path Planning Method with Direction-Constrained Encoding

Abstract

The present invention proposes a genetic algorithm continuous search path planning method with direction - limited encoding, aiming to solve the problems existing in the existing methods in the field of target search path planning, such as low search success rate, discontinuous paths, long path lengths, and large time consumption for transfer between path points. The method mainly includes the following steps: improving the gene encoding method, using the transformation direction between path points rather than directly using path points as genes, and designing an edge return mechanism; limiting the transformation direction between path points to the specified grid probability direction to avoid excessive randomness in the extension of path points; using a genetic algorithm for iterative optimization; and realizing continuous path planning for adjacent grids to reduce the number of grids spanned between path points. The simulation results show that the method of the present invention improves the search success rate, shortens the path length, reduces the overall search time, and improves the utilization rate of search resources, providing a new solution for search path planning in fields such as the manufacture of lifeboat and raft equipment.

Description

Technical Field

[0001] The present invention relates to the field of target search for unmanned systems, specifically to path planning technology for static target search, and particularly to a continuous search path planning method for a genetic algorithm with direction - limited encoding. Background Art

[0002] In the current field of marine engineering equipment manufacturing, there are numerous high - end technologies and equipment involved, such as light alloy motor housing casting or welding equipment, subsea pipeline welding equipment, drilling ships (barges), semi - submersible transport ships, etc. These equipments play an important role in the development and utilization of marine resources. At the same time, the application of auxiliary equipments such as marine hydrological and geological survey winches, general deep - sea materials and connectors, and advanced equipments such as high - frequency ground wave, S / C / X - band wave - measuring radars provides strong technical support for marine engineering. In addition, the continuous development of monitoring of marine water quality and ecological elements, and acoustic and optical measurement and detection equipments provides important means for hydrological, meteorological and water quality observations, such as hydrometeorological observation sensors, equipments and systems, buoys, moored buoys, seabed - based, mobile observation platforms, etc.

[0003] However, in the process of marine engineering equipment manufacturing and operation, how to efficiently plan the target search path has become an urgent problem to be solved. Especially in the manufacturing process of special marine engineering equipments such as polar ice - breaking ships and rock - dropping ships, and in the field of manufacturing of rescue and salvage equipments such as lifeboat rafts and inflatable life rafts, the requirements for search path planning are more stringent. In addition, the research and development of equipments such as underwater cable fault detection equipment connectors, maritime life jackets, and marine rubber and plastic lifebuoys also pose new challenges to underwater search path planning.

[0004] Target search path planning is a key issue in the fields of robot navigation, autonomous driving vehicles, and unmanned aerial vehicles. Its core task is to plan a path so that when the search unmanned boat moves along this path, the probability of finding the target is maximized as much as possible. In the field of target search path planning, improving search efficiency and success rate has always been the focus of research. With the development of technology, the path planning method for target search based on probability maps has become an important means to solve such problems. In a complex environment, the search path planning method based on probability maps has received wide attention because it can effectively handle the uncertainty of the target position. The probability map represents the environment in a rasterized manner, and each grid contains the probability information of the target's existence, thus providing a decision basis for path planning. However, the existing methods still have some deficiencies, which limit their effects in practical applications.

[0005] The problems existing in the existing methods mainly include the following aspects:

[0006] (1) Large randomness in path planning and low search success rate: The path points of the existing genetic algorithm without direction limitation extend randomly, which easily leads to the search path deviating from the optimal solution, affecting the search effect. There is a problem of low search success rate, which affects the efficiency and reliability of path planning. For example, in the papers "Research on UAV Search Path Planning Based on Improved Genetic Algorithm", "Research on Cooperative Search Path Planning Algorithm for Marine Drowning Targets", and "AUV Dynamic Target Search Algorithm Based on Improved Genetic Algorithm", the genetic algorithm gene coding methods all use path points as genes, without limiting the extension direction of path points, with large randomness and low search success rate.

[0007] (2) Poor continuity in path planning: When the traditional myopic search path planning method plans the path, there are often situations where multiple grids are spanned between path points, resulting in discontinuous paths, increasing the time consumption of transferring between path points and reducing the efficiency of the search task. For example, in Section 8.02 of "Naval Operational Research Analysis", "Naval Tactical Decision Aids", and "The Theory of Search: Optimum Distribution of Research Effort", multiple grids are spanned between path points of the myopic search method.

[0008] (3) Long path length and low search success rate: Although the path planning method of myopic search can find a feasible path, the path length is often long, resulting in insufficient utilization of search resources, and at the same time, the search success probability also needs to be improved.

[0009] Aiming at the problems that the path points of the existing genetic algorithm without direction limitation extend randomly and the search success rate is not high; the traditional myopic search path planning method spans multiple grids between path points, the path is discontinuous, and the time consumption of transferring between path points is large; the path length of the myopic search path planning method is long, and at the same time, the search success probability also needs to be improved, the present invention provides a continuous search path planning method of genetic algorithm with direction-limited coding, which improves the search success rate, shortens the path length, reduces the overall search time, and improves the utilization efficiency of search resources. Summary of the Invention

[0010] To achieve the above object, the present invention provides a continuous search path planning method of genetic algorithm with direction-limited coding, specifically including the following steps:

[0011] S1. Construct a static target model;

[0012] S2. Design a Bayesian update method for negative information of the probability map;

[0013] S3. Construct a population matrix and a coordinate summary matrix, and initialize the population;

[0014] S4. On this basis, set the parameters of the genetic algorithm and perform the iteration and optimization of the genetic algorithm;

[0015] S5. Simulate and verify the effectiveness of the proposed method.

[0016] The overall block diagram of the present invention is as Figure 1 shown.

[0017] Among them, step S1 is specifically as follows:

[0018] Construct a static target model:

[0019] Discretize the map, divide the map into R×R grids (R is a positive integer), the grids have R rows and R columns, and obtain the initial target position probability map from the pre-multiple scenarios deconstruction. The probability in each grid is the probability that the target is located in that grid. Let the grid (i, j) be the grid in the i-th row and the j-th column. The number of targets and search unmanned boats is both 1.

[0020] The initial target position probability map is as Figure 2 . In the figure, the target distribution probability is the number in the figure multiplied by 10 -4 .

[0021] Among them, step S2 is specifically as follows:

[0022] Design a Bayesian update method for the negative information of the probability map:

[0023] Take the negative information as a new condition and use Bayes' theorem to update the probability map. Bayes' formula is as follows:

[0024]

[0025] In the formula, A and B are both events, P(A k ), P(A l ) are the probabilities of events A k , A l respectively, P(B∣A k ), P(B∣A l ) are the probabilities of event B under the conditions of events A k , A l respectively, P(A k ∣B) is the probability of event A k under the condition of event B, and n is the total number of event A; Bayes' formula can reverse the conditions and conclusions of events and find the probability of the event with the reversed conditions and conclusions; the Bayes' formula combined with the specific scenario is as follows:

[0026]

[0027] Wherein, P(Target at (i,j)) is the probability that the target is in grid (i,j) in advance, P(Not detected | Target at (i,j)) is the probability of not detecting the target under the condition that the target is in grid (i,j), and P(Target at (i,j) | Not detected) is the probability that the target is in grid (i,j) under the condition of not detecting the target. P(Target at (i,j))×P(Not detected | Target at (i,j)) in advance is called the normalization factor; let the probability of search failure be:

[0028]

[0029] Then the probability of successful search is 1 - S; let the Cumulative Detection Probability (CDP) be the probability of successful search P scs :

[0030] P scs = 1 - S (4)

[0031] Use a single unmanned boat to search for a single target. The purpose of this method is to plan a path so that when the search unmanned boat moves along this path, the probability of detecting the target is as large as possible. Let t int be the time interval between adjacent path points. Assume that the unmanned boat continuously searches during movement, each path point is located at the center of the grid, and assume that the time taken by the unmanned boat to pass through each grid is equal. Then t int is both the time interval between adjacent path points and the single search time, that is, the time for the unmanned boat to pass through a grid. The single search probability P s is the search success rate within the time when the unmanned boat passes through this grid under the condition that the target is in the grid previous to the grid where the current path point is located. Every time t int passes, record the search success probability once.

[0032] Among them, step S3 is specifically as follows:

[0033] Construct a population matrix and a coordinate summary matrix, and initialize the population:

[0034] Manage data using the idea of matrices. Construct an N×M population matrix, where N is the population size, that is, the number of chromosomes in the population, and M is the number of genes in a single chromosome, that is, the chromosome length. Here, an individual corresponds to a chromosome. At the same time, construct an N×(2M + 2) coordinate summary matrix. Let t pr be the total search time, and t int be the single search time.

[0035] Take each path as a chromosome. The traditional method takes the path points on the path as genes, while this method takes the transformation method between the path points at the previous moment and the next moment as genes. Therefore,

[0036]

[0037] A schematic diagram of the relationship between the population matrix and the coordinate summary matrix is shown in Figure 3 .

[0038] In the population matrix (circular matrix), each row of this matrix represents a sequence of path point transformation methods in a path. In the coordinate summary matrix (elliptical matrix), each row of this matrix represents a sequence of path point coordinates in a path. To prevent the unmanned boat from moving to a region with a relatively small probability due to excessive randomness, the transformation method between path points at adjacent moments is designed as follows: The gene value N p in each chromosome can take any integer in [1, 2, 3, …, M p , where M p is a positive integer. If the gene value is equal to N p , then the path point transformation method is that the coordinate of the path point at the next moment is the coordinate of the grid with the N p -th largest updated probability among the eight adjacent grids around the grid pointed to by the path point coordinate at the previous moment. As Figure 4 .

[0039] Design an edge return mechanism: If the previous path point is on the edge grid of the probability map, then the next path point returns to the non-edge grid closest to the previous path point.

[0040] If there are too many path points, the chromosome length will be too long and the dimension will be too high, which will lead to an increase in the population size and the number of iterations, making the optimization speed slow. At this time, in order to reduce the chromosome length, it is evenly divided into several segments, and the remaining genes form the last segment. Each segment corresponds to a new gene, and these new genes form a new sequence in the original order as the generation chromosome. The multiple genes in each segment of the original chromosome are the same as the new genes of the corresponding generation chromosome. Use the generation chromosome to replace the original chromosome for selection, crossover, and mutation, and then transform back to the original chromosome. Using this technology can achieve chromosome dimensionality reduction.

[0041] Among them, step S4 is specifically as follows:

[0042] Set the parameters of the genetic algorithm and perform iteration and optimization of the genetic algorithm:

[0043] Set the mutation rate, crossover rate, and number of iterations.

[0044] In each iteration, first, the coordinate summary matrix is calculated from the population matrix. Secondly, the fitness value of each chromosome is calculated. Then, selection, mutation, and crossover operations are performed on the chromosomes to update the population.

[0045] After iterating several times like this, find the chromosome with the maximum fitness value in the population after several updates as the optimal individual, calculate the corresponding path as the optimal path, and calculate the search success probability corresponding to this path, and make the optimal path graph and the search success probability graph.

[0046] Take the search success probability at the last moment as the fitness function. Perform crossover and mutation operations on the chromosomes to update the population. After iterating several times, select the path corresponding to the optimal chromosome as the optimal path.

[0047] The method for calculating the coordinate summary matrix from the population matrix is as follows:

[0048] Through the transformation method between path points defined above, the corresponding path is obtained from each chromosome, and then the coordinate summary matrix is obtained. According to the coordinates of the previous path point, update the probability map through the Bayesian update of the negative information of the probability map. Find the M p grids with the largest probabilities within the eight grids around the previous path point, and determine the coordinates of the next path point according to the gene value corresponding to this place.

[0049] The method for calculating the fitness value of each chromosome is as follows:

[0050] Take the search success probability at the last moment of the path as the fitness function corresponding to the chromosome.

[0051] The methods for performing selection, mutation, and crossover operations are as follows:

[0052] The roulette wheel method is used for the selection operation. Let F(x i ) represent the fitness value of the i-th individual, then the probability of the i-th individual being selected is given by the following formula:

[0053]

[0054] To prevent the optimal chromosome from being damaged, in each round of iteration, first select the optimal chromosome to enter the offspring, and at the same time move the corresponding path initial coordinates to the corresponding positions. The remaining chromosomes are selected into the offspring according to the roulette wheel method, and their corresponding path initial coordinates are moved to the corresponding positions.

[0055] The mutation operation randomly changes the gene value to any positive integer in [1, 2, 3,..., M p at a certain mutation rate.

[0056] The crossover operation randomly selects two different chromosomes at a certain crossover rate, randomly selects the starting point and the ending point of the crossover segment, and swaps the crossover segments of the two chromosomes. As Figure 5 、 Figure 6 shown.

[0057] By comparing with the path planning of the genetic algorithm without direction limitation and the path planning of the myopic search, the present invention has the following beneficial effects:

[0058] (1) Since the method of the present invention limits the extension direction of the path points, compared with the genetic algorithm without direction limitation, in terms of the success probability of search, the success probability at the last moment increases by 23.9%, which indicates that the method of the present invention has a significant effect in improving the search success rate.

[0059] (2) Since the present invention improves the gene coding method, using the transformation direction between path points instead of directly using path points as genes, the method of the present invention effectively realizes the continuous path planning of adjacent grids, reduces the number of grids spanned between path points, avoids wasting time on the transfer between path points, realizes continuous search, thereby saving the total time of the search task and improving the utilization efficiency of search resources.

[0060] (3) Since the method of the present invention realizes the continuous path planning of adjacent grids, compared with the path planning of the myopic search, in terms of the path length, the path length is reduced by 21.8%, indicating that the method of the present invention has obvious advantages in path optimization. At the same time, in terms of the success probability of search, the success probability at the last moment increases by 15.6%, which indicates that the method of the present invention improves the success probability of search while shortening the path.

[0061] The reason why the method of the present invention has the above advantages is mainly because the genetic algorithm is used for iterative optimization, using the transformation direction between path points instead of directly using path points as genes, and limiting the extension direction of path points. This optimization strategy avoids excessive randomness in the extension of path points, ensures that path points extend in the direction with a higher probability, and improves the success probability of search. And this strategy effectively realizes the continuous path planning of adjacent grids, shortens the path length, thereby improving the search quality and efficiency, and provides a new solution for the search path planning in the fields such as the manufacture of lifeboat and raft equipment. Brief Description of the Drawings

[0062] Figure 1 It is the overall block diagram of a continuous search path planning method of a genetic algorithm with direction-limited coding;

[0063] Figure 2 It is the initial target probability map;

[0064] Figure 3Schematic diagram of the relationship between the population matrix and the coordinate summary matrix;

[0065] Figure 4 Schematic diagram of the extension direction of the current path point;

[0066] Figure 5 Schematic diagram before chromosome crossover operation;

[0067] Figure 6 Schematic diagram after chromosome crossover operation;

[0068] Figure 7 Target search planning path of the method of the present invention;

[0069] Figure 8 Search success probability graph and fitness evolution curve of the target search of the method of the present invention;

[0070] Figure 9 Target search planning path of the genetic algorithm without limited direction;

[0071] Figure 10 Search success probability graph and fitness evolution curve of the genetic algorithm without limited direction;

[0072] Figure 11 Planning path of the myopic search;

[0073] Figure 12 Search success probability of the myopic search. Detailed implementation manners

[0074] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. The following description of at least one exemplary embodiment is actually only illustrative and in no way restrictive of the present invention and its application or use. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0075] Aiming at the problems that the randomness of the extension of the path points of the existing genetic algorithm without limited direction is relatively large and the search success rate is not high; the traditional myopic search path planning method spans multiple grids between path points, the path is discontinuous, and the time consumption for transferring between path points is large; the path length of the path planning method of the myopic search is relatively long, and at the same time the search success probability also needs to be improved, the present invention provides a genetic algorithm continuous search path planning method with limited direction coding, which specifically includes the following steps:

[0076] S1. Construct a static target model;

[0077] S2. Design the Bayesian update method for negative information of the probability map;

[0078] S3. Construct the population matrix and the coordinate summary matrix, and initialize the population;

[0079] S4. On this basis, set the parameters of the genetic algorithm, and perform the iteration and optimization of the genetic algorithm;

[0080] S5. Verify the effectiveness of the proposed method through simulation.

[0081] The overall block diagram of the present invention is as Figure 1 shown.

[0082] Among them, step S1 is specifically as follows:

[0083] Construct a static target model:

[0084] Discretize the map, divide the map into R×R grids (R is a positive integer), there are R rows and R columns of grids in total. The initial target position probability map is obtained from the pre-multi-scenario deconstruction. The probability in each grid is the probability that the target is located in that grid. Let the grid (i, j) be the grid in the i-th row and the j-th column. The number of targets and search unmanned boats is both 1.

[0085] Here, take R = 20, and the initial target position probability map is as Figure 2 . In the figure, the target distribution probability is the number in the figure multiplied by 10 -4 .

[0086] Among them, step S2 is specifically as follows:

[0087] Design the Bayesian update method for negative information of the probability map:

[0088] Use the negative information as a new condition and update the probability map using Bayes' theorem. Bayes' formula is as follows:

[0089]

[0090] In the formula, A and B are both events, P(A k ), P(A l ) are the probabilities of events A k , A l respectively. P(B∣A k ), P(B∣A l ) are the probabilities of event B under the conditions of events A k , A l respectively. P(A k ∣B) is the probability of event A k under the condition of event BThe probability, where n is the total number of events A; the Bayes formula can reverse the conditions and conclusions of an event to find the probability of the event with reversed conditions and conclusions; the Bayes formula combined with a specific scenario is as follows:

[0091]

[0092] In the formula, P(Target at (i,j)) in advance is the probability that the target is in the grid (i,j) in advance, P(Not detected|Target at (i,j)) is the probability of not detecting the target under the condition that the target is in the grid (i,j), and P(Target at (i,j)|Not detected) is the probability that the target is in the grid (i,j) under the condition of not detecting the target. P(Target at (i,j)) in advance × P(Not detected|Target at (i,j)) is called the normalization factor; let the probability of search failure be:

[0093]

[0094] Then the probability of search success is 1 - S; let the Cumulative Detection Probability (CDP) be the search success probability P scs :

[0095] P scs = 1 - S (4)

[0096] Use a single unmanned boat to search for a single target. The purpose of this method is to plan a path so that when the search unmanned boat moves along this path, the probability of detecting the target is as large as possible. Let t int be the time interval between adjacent path points. Assume that the unmanned boat conducts continuous searches during movement, each path point is located at the center of the grid, and assume that the time taken by the unmanned boat to pass through each grid is equal. Then t int is both the time interval between adjacent path points and the single search time, that is, the time for the unmanned boat to pass through a grid. The single search probability P s is the search success rate within the time when the unmanned boat passes through this grid under the condition that the target is in the grid previous to the grid where the current path point is located. Every time t int passes, record the search success probability once.

[0097] Among them, step S3 is specifically as follows:

[0098] Construct a population matrix and a coordinate summary matrix, and initialize the population:

[0099] Manage data using the idea of a matrix. Construct a population matrix of N×M, where N is the population size, that is, the number of chromosomes in the population, and M is the number of genes in a single chromosome, that is, the chromosome length. Here, an individual corresponds to a chromosome. At the same time, construct a coordinate summary matrix of N×(2M + 2). Let t pr be the total search time, and t int be the single search time.

[0100] Take each path as a chromosome. The traditional method takes the path points on the path as genes, while this method takes the transformation method between the path points at the previous moment and the next moment as genes. Therefore,

[0101]

[0102] A schematic diagram of the relationship between the population matrix and the coordinate summary matrix is shown in Figure 3 .

[0103] In the population matrix (circular matrix), each row of this matrix represents a sequence of path point transformation methods in a path. In the coordinate summary matrix (elliptical matrix), each row of this matrix represents a sequence of path point coordinates in a path. To prevent the unmanned boat from moving to a region with a relatively low probability due to excessive randomness during the search, the transformation method between path points at adjacent moments is designed as follows: The gene value in each chromosome can take any integer in [1, 2, 3, 4, 5]. If the gene value is equal to 1, the path point transformation method is that the coordinate of the path point at the next moment is the coordinate of the grid with the largest updated probability among the eight adjacent grids around the grid pointed to by the path point coordinate at the previous moment in the probability map; if the gene value is equal to 2, the path point transformation method is that the coordinate of the path point at the next moment is the coordinate of the grid with the second largest updated probability among the eight adjacent grids around the grid pointed to by the path point coordinate at the previous moment in the probability map; if the gene value is equal to 3, the path point transformation method is that the coordinate of the path point at the next moment is the coordinate of the grid with the third largest updated probability among the eight adjacent grids around the grid pointed to by the path point coordinate at the previous moment in the probability map; if the gene value is equal to 4, the path point transformation method is that the coordinate of the path point at the next moment is the coordinate of the grid with the fourth largest updated probability among the eight adjacent grids around the grid pointed to by the path point coordinate at the previous moment in the probability map; if the gene value is equal to 5, the path point transformation method is that the coordinate of the path point at the next moment is the coordinate of the grid with the fifth largest updated probability among the eight adjacent grids around the grid pointed to by the path point coordinate at the previous moment in the probability map. As shown in Figure 4 .

[0104] Design an edge return mechanism: If the previous path point is at the edge grid of the probability map, then the next path point returns to the nearest non-edge grid from the previous path point.

[0105] If there are too many path points, the chromosome length will be too long and the dimension will be too high, which will lead to an increase in the population size and the number of iterations, resulting in a slowdown in the optimization speed. At this time, in order to reduce the chromosome length, it is evenly divided into several segments, and the genes of the remainder form the last segment. Each segment corresponds to a new gene, and these new genes are arranged in the original order to form a new sequence as the generation chromosome. The multiple genes in each segment of the original chromosome are the same as the new genes of the corresponding generation chromosome. The generation chromosome is used to replace the original chromosome for selection, crossover, and mutation, and then it is transformed back to the original chromosome. This technology can be used to achieve chromosome dimensionality reduction. In this embodiment, the generation chromosome is not used.

[0106] Among them, step S4 is specifically as follows:

[0107] Set the parameters of the genetic algorithm and perform the iteration and optimization of the genetic algorithm:

[0108] Set the mutation rate, crossover rate, and number of iterations.

[0109] In each iteration, first calculate the coordinate summary matrix from the population matrix. Secondly, calculate the fitness value of each chromosome. Then perform selection, mutation, and crossover operations on the chromosomes to update the population.

[0110] After iterating several times like this, find the chromosome with the maximum fitness in the updated population after several updates as the optimal individual, calculate the corresponding path as the optimal path, and calculate the search success probability corresponding to this path to create the optimal path graph and the search success probability graph.

[0111] Use the search success probability at the last moment as the fitness function. Perform crossover and mutation operations on the chromosomes to update the population. After iterating several times, select the path corresponding to the optimal chromosome as the optimal path.

[0112] The method for calculating the coordinate summary matrix from the population matrix is as follows:

[0113] Through the transformation method between path points defined above, obtain the corresponding path from each chromosome, and then obtain the coordinate summary matrix. According to the coordinates of the previous path point, update the probability map through the Bayesian update of the negative information of the probability map. Find the top five grids with the highest probability within the eight grids around the previous path point, and determine the coordinates of the next path point based on the gene values corresponding to this location.

[0114] The method for calculating the fitness value of each chromosome is as follows:

[0115] Use the search success probability at the last moment of the path as the fitness function of the corresponding chromosome.

[0116] The methods for performing selection, mutation, and crossover operations are as follows:

[0117] The selection operation uses the roulette wheel method. F(xi ) represents the fitness value of the \(i\)-th individual, and the probability that the \(i\)-th individual is selected is given by the following formula:

[0118]

[0119] To prevent the optimal chromosome from being damaged, in each iteration, the optimal chromosome is first selected to enter the offspring, and at the same time, the corresponding initial coordinates of the path are moved to the corresponding positions. The remaining chromosomes are selected into the offspring according to the roulette wheel method, and their corresponding initial coordinates of the path are moved to the corresponding positions.

[0120] The mutation operation randomly changes the gene value to any positive integer in \([1, 2, 3, 4, 5]\) at a certain mutation rate.

[0121] The crossover operation randomly selects two different chromosomes at a certain crossover rate, randomly selects the starting point and the ending point of the crossover segment, and swaps the crossover segments of the two chromosomes. As Figure 5 、 Figure 6 shown:

[0122] Among them, step S5 is specifically as follows:

[0123] Perform MATLAB simulation verification to verify the effectiveness of the above method:

[0124] Use a single unmanned boat to search for a single target. Plan a path so that when the search unmanned boat moves along this path, the probability of searching for the target is as large as possible. Assume that the unmanned boat continuously searches during movement, each path point is located at the center of the grid, and assume that the time taken by the unmanned boat to pass through each grid is equal. Let the single search time \(t\) int = 1h, the single search probability \(P\) s = 0.9, the total search time \(t\) pr = 100h. Record the search success probability \(P\) int once every time \(t\) scs . From formula (5), the chromosome length \(M = 100\). Set the mutation rate of the genetic algorithm to 0.3, the crossover rate to 0.8, and the number of iterations to 100 times.

[0125] According to the above method and parameter settings, use MATLAB for simulation.

[0126] The simulation of the method of the present invention, that is, a genetic algorithm continuous search path planning method with limited direction encoding, is as Figures 7 - 8 、Table 1 - Table 2 shown. Figures 7 - 8 are the target search planning path, search success probability graph and fitness evolution curve of the method of the present invention. The optimal individual and the optimal path are shown in Table 1 and Table 2.

[0127] Table 1 Optimal individual of the method of the present invention

[0128]

[0129] Its optimal value, that is, the search success probability at the last moment, is 0.867.

[0130] Table 2 Optimal Path of the Method of the Present Invention

[0131]

[0132] Its path length is 119.468 grid units. Among them, the abscissa is the row number of the grid where the path point is located, and the ordinate is the column number of the grid where the path point is located.

[0133] The simulation of the continuous search path planning method of the genetic algorithm without a defined direction is as Figures 9 - 10 shown in Tables 3 - 4. Figures 9 - 10 It is the target search planning path, search success probability graph, and fitness evolution curve of the genetic algorithm without a defined direction. The optimal individual and optimal path are shown in Tables 3 and 4.

[0134] Table 3 Optimal Individual of the Genetic Algorithm without a Defined Direction

[0135]

[0136] Its optimal value, that is, the search success probability at the last moment, is 0.700.

[0137] Table 4 Optimal Path of the Genetic Algorithm without a Defined Direction

[0138]

[0139]

[0140] Its path length is 121.539 grid units. Among them, the abscissa is the row number of the grid where the path point is located, and the ordinate is the column number of the grid where the path point is located.

[0141] The path planning method of the myopic search is as Figures 11 - 12 shown in Table 5.

[0142] Table 5 Optimal Path of the Myopic Search

[0143]

[0144] Its path length is 152.678 grid units, and its search success probability at the last moment is 0.750.

[0145] The method of the present invention is respectively compared with the path planning of the genetic algorithm without a defined direction and the path planning of the myopic search. Figure 8 With Figure 10In comparison, the search success probability at the last moment of the method of the present invention is 0.867, and the search success probability at the last moment of the genetic algorithm without direction limitation is 0.700, indicating that compared with the genetic algorithm without direction limitation in search path planning, the search success probability at the last moment of the method of the present invention has increased by 23.9%. This is because the method of the present invention limits the extension direction of the path points, that is, the direction with the top M p large grid probabilities. This can avoid excessive randomness in the extension of path points and ensure that path points extend in directions with probabilities not too small. Figure 7 In comparison with Figure 11 It can be seen that the method of the present invention realizes continuous path planning for adjacent grids. The grids where the previous path point and the next path point of the method of the present invention are located are adjacent, while there are generally multiple grids spanned between the previous path point and the next path point of the near-sighted search path planning method, indicating that the method of the present invention reduces the number of grids spanned between path points, avoids consuming time in the transfer between path points, realizes continuous search, saves the total time of the search task, and improves the utilization efficiency of search resources. This is because the method of the present invention improves the gene coding method, using the transformation direction between path points instead of directly using path points as genes. Moreover, the path length of the method of the present invention is 119.468 grid units respectively, while the path length of the near-sighted search path planning is 152.678 grid units, indicating that compared with the near-sighted search path planning, the path length of the method of the present invention has decreased by 21.8%. This is because the method of the present invention realizes continuous path planning for adjacent grids. Figure 8 In comparison with Figure 12 It can be seen that the search success probability at the last moment of the method of the present invention is 0.867, and the search success probability at the last moment of the near-sighted search path planning is 0.750, indicating that compared with the near-sighted search path planning, the search success probability at the last moment of the method of the present invention has increased by 15.6%. This is because this method uses a genetic algorithm for optimization and limits the extension direction of path points.

[0146] The method of the present invention realizes continuous path planning for adjacent grids by using genetic algorithm iterative optimization, limiting the extension direction of path points, and improving the gene coding method of the genetic algorithm, thereby improving the search success rate, shortening the path length, reducing the overall search time, and improving the utilization efficiency of search resources, providing a new solution for search path planning in fields such as lifeboat and raft equipment manufacturing.

[0147] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them: Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features: And these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A continuous search path planning method for a genetic algorithm with direction-limited coding, characterized in that, It includes the following steps: S1. Construct a static target model; S2. Design a Bayesian update method for negative information of the probability map; S3. Construct a population matrix and a coordinate summary matrix, and initialize the population; S4. On this basis, set the parameters of the genetic algorithm, and perform iteration and optimization of the genetic algorithm; S5. Simulate and verify the effectiveness of the proposed method; Among them, step S1 is specifically as follows: Construct a static target model: Discretize the map, divide the map into R×R grids (R is a positive integer), the grids have R rows and R columns, and obtain the initial target position probability map from the pre-multi-scenario deconstruction. The probability in each grid is the probability that the target is in this grid; let the grid (i,j) be the grid in the i-th row and the j-th column; the number of targets and search unmanned boats is 1; Among them, step S2 is specifically as follows: Design a Bayesian update method for negative information of the probability map: Take the negative information as a new condition and use Bayes' theorem to update the probability map; the Bayes formula is as follows: where A and B are both events, P(A k ), P(A l ) are the probabilities of events A k and A l respectively, P(B∣A k ), P(B∣A l ) are the probabilities of event B under the conditions of events A k and A l respectively, P(A k ∣B) is the probability of event A k under the condition of event B, n is the total number of all cases of event A; the Bayes' formula can reverse the conditions and conclusions of events to find the probability of the event with reversed conditions and conclusions; the Bayes' formula combined with a specific scenario is as follows: Wherein, P(Target at (i, j)) in advance is the probability that the target is in the grid (i, j) in advance, P(Not detected | Target at (i, j)) is the probability of not detecting the target under the condition that the target is in the grid (i, j), and P(Target at (i, j) | Not detected) is the probability that the target is in the grid (i, j) under the condition of not detecting the target; is called the normalization factor; Let the probability of search failure be: The probability of a successful search is 1 - S; let the Cumulative Detection Probability (CDP) be the probability of a successful search P scs : P scs = 1 - S(4) Use a single unmanned boat to search for a single target; the purpose of this method is to plan a path so that when the search unmanned boat moves along this path, the probability of finding the target is maximized as much as possible; let t int be the time interval between adjacent path points. Assume that the unmanned boat continuously searches during movement. Each path point is located at the center of the grid, and assume that the time taken by the unmanned boat to pass through each grid is equal. Then t int is not only the time interval between adjacent path points but also the single search time, that is, the time for the unmanned boat to pass through a grid; the single search probability P s is the search success rate within the time when the unmanned boat passes through this grid under the condition that the target is in the grid previous to the grid where the current path point is located; every time t int passes, record the search success probability once; Among them, step S3 is specifically as follows: Construct a population matrix and a coordinate summary matrix, and initialize the population: Manage data using the idea of a matrix; construct a population matrix of N×M, where N is the population size, that is, the number of chromosomes in the population, and M is the number of genes in a single chromosome, that is, the chromosome length; here, an individual corresponds to a chromosome; at the same time, construct a coordinate summary matrix of N×(2M + 2); let t pr be the total search time, and t int be the single search time; Take each path as a chromosome; the traditional method takes the path points on the path as genes, and this method takes the transformation method between the previous moment and the next moment path points as genes; Each row of the population matrix represents a sequence of path point transformation methods in a path; each row of the coordinate summary matrix represents a sequence of path point coordinates in a path; to prevent the search unmanned boat from moving to an area where the probability value is lower than the preset threshold due to excessive randomness, the transformation method between adjacent path points is designed as follows: the gene value N in each chromosome p can take any integer in [1, 2, 3, …, M p , where M p is a positive integer; if the gene value is equal to N p , then the path point transformation method is that the coordinates of the path point at the next moment are the coordinates of the grid with the N p th largest updated probability in the eight adjacent grids around the grid pointed to by the path point coordinates at the previous moment; Design an edge return mechanism: if the previous path point is in the edge grid of the probability map, then the next path point returns to the nearest non-edge grid from the previous path point; If there are too many path points, the chromosome length will be too long and the dimension will be too high, which will lead to an increase in the population size and the number of iterations, resulting in a slow optimization speed; at this time, in order to reduce the chromosome length, it is evenly divided into several segments, and the remaining genes form the last segment. Each segment corresponds to a new gene, and these new genes form a new sequence in the original order as the generation chromosome; the multiple genes of each segment of the original chromosome are the same as the new genes of the corresponding generation chromosome; use the generation chromosome to replace the original chromosome for selection, crossover, and mutation, and then transform back to the original chromosome; using this technology can achieve chromosome dimension reduction; Among them, step S4 is specifically as follows: Set the parameters of the genetic algorithm, and perform iteration and optimization of the genetic algorithm: Set the mutation rate, crossover rate, and number of iterations; In each iteration, first calculate the coordinate summary matrix from the population matrix; secondly, calculate the fitness value of each chromosome; then perform selection, mutation, and crossover operations on the chromosomes to update the population; After iterating several times like this, find the chromosome with the maximum fitness value in the updated population after several updates as the optimal individual, calculate the corresponding path as the optimal path, and calculate the search success probability corresponding to this path, and make the optimal path map and search success probability map; Take the search success probability at the last moment as the fitness function; perform crossover and mutation operations on the chromosomes to update the population. After iterating several times, select the path corresponding to the optimal chromosome as the optimal path; The method for calculating the coordinate summary matrix from the population matrix is as follows: Through the transformation method between the path points defined in step S3, the corresponding path is obtained from each chromosome, and then the coordinate summary matrix is obtained; according to the coordinates of the previous path point, the probability map is updated through the Bayesian update of the negative information of the probability map; within the eight grids around the previous path point, find the grids with the top M p large grids, and the coordinates of the next path point are determined by the gene value corresponding to this location; The method for calculating the fitness value of each chromosome is as follows: Take the search success probability at the last moment of the path as the fitness function of the corresponding chromosome; The methods for performing selection and mutation operations are as follows: The selection operation adopts the roulette wheel method; to prevent the best chromosome from being damaged, in each iteration, the best chromosome is first selected to enter the offspring, and at the same time, the corresponding path initial coordinates are moved to the corresponding positions; the remaining chromosomes are selected into the offspring according to the roulette wheel method, and their corresponding path initial coordinates are moved to the corresponding positions; The mutation operation randomly changes the gene value to any positive integer in [1, 2, 3, …, M p at a certain mutation rate.

Images