Improved ga-sa path planning method for water unmanned surface vehicle for offshore wind farm inspection

Abstract

The application relates to a water surface unmanned ship improved GA-SA path planning method for offshore wind farm inspection, and relates to a water surface unmanned ship path planning method.The application aims to solve the problems of low efficiency and low precision of an existing algorithm.The process is as follows: one, initializing algorithm parameters;two, performing chromosome coding of feasible solutions, and setting a fitness value function;three, taking the highest fitness value as the optimal individual;four, judging whether the convergence condition is met, if yes, outputting the current optimal individual, otherwise, entering five;five, taking the current optimal value as an initial solution of the SA algorithm, and randomly obtaining an updated solution;judging the energy change of the updated solution;if the energy change is less than 0, the updated solution is taken as the latest solution;otherwise, the updated solution is reserved according to a probability;six, judging whether the convergence condition is met, if no, repeating five;if yes, returning to three to replace the optimal individual, and repeatedly executing four to six until the convergence condition is met.The application is used in the field of path planning of water surface unmanned ships.

Description

Technical Field

[0001] This invention relates to a path planning method for unmanned surface vessels. Background Technology

[0002] Unmanned surface vehicles (USVs) are one type of unmanned platform, but their development is later than that of drones, unmanned vehicles, and unmanned underwater vehicles. Compared to manned ships, USVs, with their typical advantages of being unmanned, small in size, and high in speed, can more efficiently complete some civilian and scientific research tasks, such as marine resource exploration, marine geographic information collection, nautical charting, and maritime search and rescue. They can also perform various military tasks, such as maritime intelligence gathering, monitoring of key sea areas, mine clearance operations, anti-submarine detection, and counter-terrorism operations. [1-3] ([1]Li M., He Y., Ma Y., et al. Design and implementation of a new jet-boat based unmanned surface vehicle[C]. Proceedings of International Conference on Automatic Control &Artificial Intelligence, Xiamen, China, 2012: 768-771. )([2]Villa J L., PaezJ., Quintero C., et al. Design and control of an unmanned surface vehicle forenvironmental monitoring applications[C]. Proceedings of Robotics &Automation, Bogota, Colombia, 2017: 1-5. )([3]Specht C., Switalski E., SpechtM. Application of an autonomous / unmanned survey vessel (ASV / USV) and bathymetric measurements[J]. Polish Marine Research, 2017, 24(3): 36-44.), such as Figure 1 As shown.

[0003] Global path planning refers to the planning of global paths based on the known environment and certain evaluation indicators and collision avoidance criteria. However, some traditional algorithms, while simple to compute and easy to operate, are inefficient and produce low-precision results. On the other hand, while advanced algorithms such as biomimetic algorithms greatly improve computational accuracy, they are more computationally cumbersome.

[0004] Therefore, this invention addresses the global path planning problem of unmanned surface vessels (USVs) for offshore wind farm inspection, and proposes an improved GA-SA path planning method for USVs for offshore wind farm inspection. Summary of the Invention

[0005] The purpose of this invention is to address the problems that existing methods cannot find the shortest path in the global path of unmanned surface vessels (USVs) for offshore wind farm inspection, and that the efficiency of determining the optimal path globally is low, and the matching accuracy between the determined optimal path and the actual optimal path is low. Therefore, this invention proposes an improved GA-SA path planning method for USVs used in offshore wind farm inspection.

[0006] The specific process of an improved GA-SA path planning method for unmanned surface vessels used for offshore wind farm inspection is as follows:

[0007] Step 1: Initialize the parameters of the improved genetic simulated annealing algorithm, including population size, adaptive crossover, mutation operator, initial temperature, cooling coefficient, Markov chain length, and cutoff temperature.

[0008] Step 2: Based on the characteristics of the solution, perform chromosome encoding and set the fitness value function;

[0009] Step 3: Decode all chromosomes and calculate the fitness value of all individuals, then select the individual with the highest current fitness as the optimal individual;

[0010] Step 4: Determine if the current population meets the convergence condition. If it does, the algorithm ends and outputs the current best individual. If it does not meet the condition, proceed to the evolutionary stage and perform selection, crossover, and mutation operations.

[0011] Step 5: Use the optimal value obtained from the evolved new generation population as the initial solution for the simulated annealing algorithm, and then execute Step 6;

[0012] The optimal value is the individual with the highest current fitness.

[0013] Step 6: Generate a random perturbation in the neighborhood of the optimal value obtained from the new generation population in Step 5 to obtain an updated solution; determine the energy change between the updated solution and the optimal solution.

[0014] If the energy of the updated solution is less than the energy of the optimal solution, then the updated solution replaces the optimal solution, and the latest solution is obtained.

[0015] If the energy of the updated solution is greater than or equal to the energy of the optimal solution, then proceed according to probability. The updated solution is retained, and the retained updated solution is the current latest solution;

[0016] In the formula The value of the change in energy. The current temperature;

[0017] Step 7: Determine whether the latest solution meets the convergence condition of the simulated annealing algorithm. If not, repeat step 6. If it does, return the latest solution to step 3 to replace the individual with the highest fitness in step 3, and repeat steps 4 to 7 until the convergence condition of the genetic algorithm is met.

[0018] The beneficial effects of this invention are as follows:

[0019] This invention first introduces the genetic algorithm and simulated annealing algorithms, analyzing their advantages and disadvantages respectively. Then, based on these advantages and disadvantages, improvements are made to both the genetic algorithm and the simulated annealing algorithm. To address the problem that the fixed crossover and mutation operators in the genetic algorithm are not applicable to the entire evolutionary process, an adaptive crossover and mutation operator is proposed, and its effectiveness and superiority are verified through simulation experiments of finding the function's extremum. Secondly, to address the slow convergence speed of the simulated annealing algorithm in the early stages, an adaptive cooling coefficient is proposed to accelerate the algorithm's convergence in the early stages, and its effectiveness and superiority are verified through simulation experiments of solving the TSP problem. Finally, simulation experiments were conducted using the improved genetic algorithm to perform USV full path planning in offshore wind farm problems, and the results were compared with those obtained using traditional genetic algorithms and traditional simulated annealing algorithms alone. The improved method yielded more efficient path planning results, thus verifying its superiority. This invention addresses the problems that while existing traditional algorithms are computationally simple and easy to operate, they are inefficient and have low accuracy, as well as that while existing advanced algorithms such as biomimetic algorithms have greatly improved accuracy, they are more computationally cumbersome.

[0020] Table 2 shows the results obtained by the three algorithms. The traditional genetic algorithm yields a path length of 16.95, while the traditional simulated annealing algorithm yields 17.37. This indicates that the genetic algorithm is slightly better than the simulated annealing algorithm, thanks to the genetic algorithm's strong global space search capability, although the optimization margin is only 2%. The improved genetic simulated annealing algorithm, however, yields a path length of 15.51, representing an 8% improvement over the genetic algorithm. This demonstrates that the improved GA-SA algorithm proposed in this invention significantly improves the accuracy compared to the traditional algorithm. Based on the above analysis, the improved genetic simulated annealing algorithm effectively complements the strengths and weaknesses of the two algorithms, ultimately improving the algorithm's accuracy. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of an unmanned surface vessel.

[0022] Figure 2 for Function graph;

[0023] Figure 3a This is the initial position map of the simulation solution of the traditional genetic algorithm;

[0024] Figure 3b This is a map showing the final location of the simulation solution from a traditional genetic algorithm.

[0025] Figure 3c A simulation of the optimal function value variation trend of a traditional genetic algorithm;

[0026] Figure 4a To improve the initial position map of the simulated solution of the genetic algorithm;

[0027] Figure 4b To improve the final location map of the simulated solution of the genetic algorithm;

[0028] Figure 4c To improve the simulation of the optimal function value change trend graph in the genetic algorithm;

[0029] Figure 5a The final roadmap for simulating the traditional simulated annealing algorithm;

[0030] Figure 5b This is a simulation trend chart of the route length variation using the traditional simulated annealing algorithm.

[0031] Figure 6a To improve the simulated annealing algorithm, a final simulation roadmap was developed.

[0032] Figure 6b To improve the simulated annealing algorithm, a simulation path length variation trend diagram was generated.

[0033] Figure 7a The image shows the simulation results of the traditional genetic algorithm.

[0034] Figure 7b The image shows the simulation results of the traditional simulated annealing algorithm.

[0035] Figure 8 The simulation results are shown in the figure to improve the algorithm. Detailed Implementation

[0036] Specific Implementation Method 1: The specific process of the improved GA-SA path planning method for unmanned surface vessels used in offshore wind farm inspections in this implementation method is as follows:

[0037] Genetic Algorithm

[0038] Genetic Algorithm (GA) is based on Darwin's theory of evolution and incorporates the principle of "survival of the fittest" in the process of biological evolution to find the optimal value in the reproduction and evolution of a population. GA determines the fitness of individuals by calculating the fitness value of all individuals in the population and completes the "survival of the fittest" process through selection, crossover, and mutation operations. Its main advantages are as follows:

[0039] (1) It has a strong global optimization capability;

[0040] (2) During the optimization process, individuals in the population are eliminated and retained by probability, without the need to formulate different evolutionary rules according to different population types. The calculation is simple and the adaptive ability is strong.

[0041] The main components of a genetic algorithm are: chromosome encoding, fitness, selection, crossover, and mutation.

[0042] Simulated annealing algorithm

[0043] Simulated Annealing (SA) is derived from the cooling process of an object. When the initial temperature of an object is high, the object is in an unstable state with a large internal energy. However, as the temperature of the object gradually decreases, the internal energy decreases, and the object begins to stabilize until the temperature of the object drops to the minimum value, the internal energy is reduced to the minimum, and the object enters an equilibrium state.

[0044] First, a relatively high initial temperature is established, which serves as the initial value for the simulated annealing algorithm. Then, the temperature is gradually decreased at a certain rate until the system reaches an equilibrium state, yielding a stable solution. However, this may be a local optimum. To improve the accuracy of the simulated annealing algorithm, it accepts a random new solution with Metappolis probability, thus escaping local optima and searching for the global optimum.

[0045] In order to achieve global path planning control of unmanned surface vessels using a global path planning controller based on the improved GA-SA algorithm, the goal of this invention is to design a global path planning controller that enables unmanned surface vessels to plan an ideal path before the start of an engineering task.

[0046] Improvements to genetic algorithms

[0047] The greatest advantage of genetic algorithms is that we don't need to design a detailed process for finding the optimal solution; we can simply eliminate underperforming individuals. However, genetic algorithms have poor local search capabilities, leading to premature convergence. Furthermore, traditional genetic algorithms often use fixed crossover and mutation operators. Even after multiple adjustments through experiments, it's difficult to guarantee that these values ​​will be optimal in the next experiment, thus affecting the algorithm's efficiency and accuracy. Among the main components of genetic algorithms, the crossover and mutation operators are two key factors determining their optimization performance. [4] ([4] ZhangJ, Chung SH, Lo WL. Clustering-based adaptive crossover and mutation probabilities for genetic algorithms [J]. IEEE Transactions on Evolutionary Computation, 2007, 11(3): 326-335.).

[0048] The quality of the crossover operator determines the diversity of individuals in the population during evolution, and the crossover process plays a dominant role in genetic algorithms. The mutation operator, on the other hand, plays a decisive role in the global search capability of the genetic algorithm; however, improperly configured mutation operators can disrupt the healthy diversity of the population.

[0049] In traditional genetic algorithms for solving optimization problems, the appropriate values ​​for crossover and mutation operators are often selected through several experiments, or determined based on previous experience. However, since genetic algorithms work by starting with a global random population and converging to the final optimal value based on fitness, this approach is flawed.

[0050] Adaptive crossover, mutation operator

[0051] Based on the above analysis, this paper proposes an adaptive crossover and mutation operator, which can dynamically adjust its value during population evolution to better adapt to each generation's evolution. If the difference between the average fitness value and the optimal fitness value of the current population is large, it can be considered that the current population is relatively dispersed, i.e., possessing rich species diversity. According to the principle of genetic algorithms, in this case, the value of the crossover operator should be increased to allow the current population to more fully perform the crossover operation and reproduce more high-quality individuals. At the same time, the value of the mutation operator should be slightly decreased to reduce the probability of high-quality individuals being destroyed, thereby improving the convergence speed of the genetic algorithm and enabling it to converge as quickly as possible. Conversely, the opposite is true. The adaptive crossover and genetic operator is derived as follows:

[0052] Adaptive Crossover:

[0053]

[0054] In the formula For adaptive coefficients, This represents the current average fitness value of the population. This represents the current maximum fitness value for the race. This corresponds to the current situation of population dispersion;

[0055] Mutation operator:

[0056]

[0057] In the formula These are adaptive coefficients;

[0058] Improved Genetic Algorithm Simulation Experiment

[0059] To verify the superiority of the improved genetic algorithm, this invention conducts simulation experiments on the algorithm. The most common problem of finding the extremum of a function is selected to verify the algorithm.

[0060] The test function is selected as The domain is set to Find the maximum value of the function within its domain. The graph of the function is shown below. Figure 2 As shown:

[0061] After several initial adjustments, the adaptive coefficients in the adaptive crossover and mutation operators were determined. , The initial number of solutions is set to 20, meaning the initial population size is 20. This paper uses a binary encoding sequence because a binary number of a certain length can represent a floating-point number of a certain precision. This invention requires the solution to have a precision of four decimal places because... Therefore, the binary string used for encoding must be at least 17 bits long. The method for converting the binary string to the corresponding floating-point number within the specified range is as follows:

[0062] Convert the binary number to its corresponding decimal number:

[0063]

[0064] Then convert the decimal number to a floating-point number within the corresponding range:

[0065]

[0066] The selection rule combines roulette wheel selection and optimal retention strategies. Roulette wheel selection first calculates the ratio of each individual's fitness value to the sum of the fitness values ​​of all individuals in the population, and then uses this ratio as the probability of whether that individual will enter the next generation. This method maintains good population diversity, but it also loses high-fitness individuals while retaining low-fitness individuals. Therefore, this paper also employs the optimal selection strategy. During the selection process, the roulette wheel selection method is first performed, and then the structure of the individual with the highest fitness in the current population is completely replicated into the next generation.

[0067] The crossover rule uses uniform crossover, which exchanges genes at every gene position of two individuals with a certain crossover probability, thereby generating two new individuals.

[0068] The mutation rule uses basic bit mutation, which involves randomly selecting one or more gene positions in the individual's encoded string and performing a mutation operation with a mutation probability. In genetic algorithms using binary encoding, the mutation operation involves selecting a random gene position and performing a mutation operation with a mutation probability. Complete the conversion between 0 and 1.

[0069] Simulation experiments were conducted in MATLAB using both traditional and improved genetic algorithms. The simulation experiment of the traditional genetic algorithm was conducted first.

[0070] Based on simulation results Figure 3a , 3b As can be seen from 3c, the traditional genetic algorithm solves for the maximum value of the function. .Depend on Figure 3b This is clearly an incorrect answer. The reason is that the fixed values ​​of the crossover and mutation operators cannot be dynamically adjusted during the evolutionary process, which has a negative impact on certain generations of the population. This causes the population to "prematurely converge," preventing it from further searching for the global optimum and directly affecting the accuracy of the algorithm.

[0071] Adaptive crossover and mutation operators were added to the traditional genetic algorithm, and simulation experiments were conducted.

[0072] Based on simulation results Figure 4a , 4b As can be seen from 4c, the maximum value of the function is found by the traditional genetic algorithm. .Depend on Figure 4a It can be seen that the initial solutions are very scattered, and after evolution, the final positions are determined by... Figure 4bAs shown, after the algorithm converges, the vast majority of the population clusters near the optimal solution, without exhibiting the "premature convergence" phenomenon seen in traditional genetic algorithms. This is because the values ​​of the crossover and mutation operators change with the current population's fitness concentration during evolution, allowing them to quickly escape extreme regions and guiding the entire population towards higher global fitness values. Figure 4c It can be seen that the improved genetic algorithm approaches the optimal solution around the 15th generation. Around the 23rd generation, a bad solution is generated due to mutation, but it is quickly eliminated. The subsequent evolution process tends to be stable, always focusing on the vicinity of the optimal solution until the algorithm converges.

[0073] Through the simulation experiments of the traditional genetic algorithm and the improved genetic algorithm with adaptive crossover and mutation operators, it is easy to see that the improved genetic algorithm reinforcement learning can effectively avoid the problem of "premature convergence", improve the computational efficiency of the algorithm, and enhance the accuracy of the algorithm.

[0074] Improvements to the simulated annealing algorithm

[0075] The advantage of simulated annealing lies in its strong local search capability. However, as explained in the principle of simulated annealing above, the solution generated at each step is randomly generated, lacking an understanding of the global environment. Therefore, it is prone to getting stuck at local optima. [5,6] ([5] Tao Zhongben, Lei Zhubing, Li Chunguang, et al. Path planning for handling robots based on improved simulated annealing algorithm [J]. Computer Measurement & Control, 2018, v.26; No.238(07):191-194.) ([6] He Jinfu, Fu Qiang, Wang Haodong. Improved simulated annealing algorithm for solving TSP problem [J]. Computer Era, 2019(7):47-50.) In addition, in order to ensure that the simulated annealing algorithm has high accuracy, the starting temperature is usually set higher and the cooling rate is slower, but this results in the initial cooling rate being too slow. [7] ([7] Li Jiangwei, Xu Lunhui. Research on the combination of annealing algorithm and neural network algorithm in path planning[J]. Automation and Instrumentation, 2017(11):11-14+36.), thus affecting the convergence speed of the entire algorithm. The core of the performance of simulated annealing algorithm is the selection of relevant parameters. Therefore, this part mainly studies the main parameter selection rules in simulated annealing algorithm, and proposes an adaptive cooling acceleration coefficient to address the problem of slow cooling speed in the early stage of simulated annealing algorithm.

[0076] Key parameter settings for simulated annealing algorithm

[0077] The simulated annealing algorithm mainly includes four key parameters: initial temperature, cooling coefficient, Markov chain length, and cutoff temperature.

[0078] initial temperature

[0079] The global search capability of the simulated annealing algorithm largely depends on the initial temperature setting. A higher initial temperature increases the probability of obtaining a higher-quality solution, but also reduces the convergence speed. Therefore, the initial temperature setting must balance the trade-off between solution accuracy and convergence speed. In experiments, the required solution accuracy needs to be determined based on the problem's accuracy requirements. If further increasing the initial temperature does not significantly improve the solution accuracy, then the lower temperature at which the required accuracy was first achieved can be chosen as the initial temperature value.

[0080] Cooling coefficient

[0081] The convergence process of the simulated annealing algorithm is achieved by gradually reducing the system temperature to room temperature. Currently, there are two main cooling methods: the first uses a classic cooling function, and the second uses a cooling coefficient to achieve a proportional temperature decrease. This paper adopts the second cooling method, where the temperature decrease rate is controlled by the cooling coefficient. Typically, a slower cooling rate is needed to ensure higher solution accuracy, but this also affects the algorithm's convergence speed. To accelerate the convergence speed while maintaining a certain level of solution accuracy, this invention proposes an adaptive cooling coefficient.

[0082] Markov chain length

[0083] The convergence of the simulated annealing algorithm requires that the solutions generated by the algorithm tend to reach equilibrium at each temperature obtained through iteration. This necessitates a sufficiently large number of solutions at each temperature. Currently, a common method for determining the number of solutions at each temperature is the Metropolis criterion, with the core parameter being the Markov chain length. A larger value indicates that the simulated annealing algorithm can iterate more times at the current temperature, thus obtaining more solutions at that temperature. Furthermore, since the algorithm receives new solutions with a certain probability during iteration at the current temperature, the value of the Markov chain length also determines the algorithm's search capability in the global space. Typically, the Markov chain length is related to the number of possible solutions, *m*, and is either a fixed value or a mapping relationship is established between *m* and the chain length. The typical range is [range missing in original text]. .

[0084] Cut-off temperature

[0085] The cutoff temperature determines the convergence condition of the algorithm; the algorithm terminates when the system temperature drops to the cutoff temperature. Typically, to ensure sufficient time to find a better solution, the cutoff temperature is set close to 0. This paper sets the cutoff temperature to 0.001.

[0086] Adaptive cooling coefficient

[0087] In the cooling process of the simulated annealing algorithm, the second cooling function from the previous section is used. The system temperature will decrease linearly. However, because the initial temperature is often designed to be relatively high, the difference between it and the cutoff temperature is large. Therefore, the iteration is very slow in the early stages of the algorithm. In practical applications, however, a large proportion of the randomly selected solutions in the early stages of iteration are of poor quality. Therefore, the temperature decrease rate can be accelerated in the early stages of the algorithm, and then reduced in the later stages to obtain a more accurate solution. Therefore, this paper proposes an adaptive cooling acceleration coefficient as shown in the following formula:

[0088]

[0089] By adopting the adaptive cooling coefficient proposed in this paper, the overall convergence time of the system is significantly reduced, allowing for a more appropriate increase in the Markov chain length. This enhances the ability of simulated annealing to perform a global search and improves the accuracy of the solution. Substituting this into the cooling function, we obtain the adaptive cooling function:

[0090]

[0091] Improved Simulated Annealing Algorithm Simulation Experiment

[0092] To verify the effectiveness and superiority of the improved simulated annealing algorithm, the Traveling Salesman Problem (TSP) was used for validation. The TSP is a classic combinatorial optimization problem. The TSP involves a traveling salesman who wants to visit... Given a city, the traveler needs to choose a route that passes through each city, with the constraint that each city can only be visited once and the destination must be the city from which the traveler departs. The length of the route is used as the evaluation metric; the shorter the selected route, the higher the evaluation metric.

[0093] Following the method for setting key parameters of the simulated annealing algorithm described above, the initial temperature... Cooling coefficient Markov length Cut-off temperature First, a MATLAB simulation was performed using the classic simulated annealing algorithm. The number of cities was set to 20, and the location of each city was randomly generated, as shown in Table 1.

[0094]

[0095] Then, a simulation experiment was conducted in MATLAB, and the results are as follows: Figure 5a , 5b 6a, 6b:

[0096] Depend on Figure 5b and Figure 6b It can be seen that the improved simulated annealing algorithm exhibits a steeper descent trend in the early stages. The traditional simulated annealing algorithm reaches approximately 20% of the optimal solution by the 450th iteration, while the improved simulated annealing algorithm reaches approximately 20% of the optimal solution by the 260th iteration. Therefore, the improved simulated annealing algorithm has a faster convergence speed than the traditional simulated annealing algorithm. Figure 6b It can be seen that after the 415th iteration, the change in the optimal solution for each update is within 0.5%, indicating that the system tends to stabilize, and the optimal path length obtained is 388.1312. Due to the increase in the Markov chain length, the improved genetic simulated annealing algorithm has a stronger global space search capability and a greater probability of finding a better value. At the same time, due to the application of the adaptive cooling coefficient, the overall convergence time of the algorithm does not increase with the increase in the Markov chain length, but rather improves the convergence speed.

[0097] Through the comparative experiments of the traditional simulated annealing algorithm and the improved simulated annealing algorithm, it is easy to see that the improved simulated annealing algorithm can improve the solution accuracy and speed up the convergence speed, thereby improving the computational efficiency of the algorithm when solving combinatorial optimization problems.

[0098] Global path planning based on the improved GA-SA algorithm

[0099] The preceding sections addressed some shortcomings of traditional genetic algorithms and simulated annealing algorithms, and simulation experiments verified the effectiveness and superiority of the improved algorithms. However, the advantage of genetic algorithms lies in their strong global spatial search capability, while the greatest advantage of simulated annealing algorithms is their strong local spatial search capability. Therefore, this invention patents combine the two algorithms and apply them to global path planning for USVs in offshore wind farm inspection problems.

[0100] Improved computational flow of GA-SA algorithm

[0101] The improved genetic simulated annealing algorithm proposed in this invention uses the traditional genetic algorithm as its basic framework. After the genetic algorithm generates a new population, it is then incorporated into the simulated annealing algorithm for local optimization, thereby improving the accuracy of the overall algorithm.

[0102] The process of improving the genetic simulated annealing algorithm is as follows:

[0103] Step 1: Initialize the parameters of the improved genetic simulated annealing algorithm, including population size, adaptive crossover, mutation operator, initial temperature, cooling coefficient, Markov chain length, and cutoff temperature.

[0104] Step 2: Based on the characteristics of the problem solutions (i.e., different encoding methods are selected for different types of feasible solutions), perform chromosome encoding operations (each solution to the optimization problem becomes a chromosome in the genetic algorithm, and different chromosomes are distinguished by different encoding methods. The encoding method represents the operation method of the genetic algorithm and also affects the efficiency of the algorithm). Set the fitness value function (the quality of each individual is judged by the fitness value, and then individuals are eliminated according to the fitness value. In the process of population evolution, the genetic algorithm can find the optimal solution based on the fitness of individuals, guiding the algorithm to always move towards the direction with higher fitness. Generally, when solving practical problems, the fitness function in the genetic algorithm needs to be determined according to the objective function of the practical problem; for example, if we want to find a shortest path, then this fitness function is the formula for calculating the path length).

[0105] Step 3: Decode all chromosomes and calculate the fitness value of all individuals (the quality of each individual is determined by its fitness value, and then individuals are eliminated based on their fitness. During population evolution, the genetic algorithm can find the optimal solution based on the fitness of individuals, guiding the algorithm to always move towards higher fitness. Generally, when solving practical problems, the fitness function in the genetic algorithm needs to be determined based on the objective function of the problem; for example, if we want to find the shortest path, then this fitness function is the formula for calculating the path length). Select the individual with the highest current fitness as the optimal individual.

[0106] Step 4: Determine whether the current population (the optimal individual is the individual with the highest fitness value in the current population) meets the convergence condition. If the convergence condition is met, the algorithm ends and the current optimal individual is output; if the condition is not met, the evolutionary stage is entered, and selection, crossover, and mutation operations are performed.

[0107] Selection: Selection is the process of choosing offspring individuals from the parent population according to certain rules, and then passing them on to the next generation. The main principle is that, based on the fitness value calculated above, individuals with higher fitness values ​​have a higher probability of being passed on to the next generation.

[0108] Crossover: To enable the genetic algorithm's evolutionary process to generate genetic genes that are more adapted to the environment and enhance population diversity, the algorithm employs a crossover operation. This refers to the exchange of gene segments between two chromosomes according to certain rules, thereby forming two new individuals.

[0109] Mutation: Mutation occurs when genes mutate. After mutation, a population can produce entirely new gene forms. These new genes may enable organisms to have higher adaptability, which can then be inherited by the next generation. Therefore, mutation accelerates the evolution of the current population.

[0110] Step 5: Use the optimal value obtained from the evolved new generation population as the initial solution for the simulated annealing algorithm, and then execute Step 6;

[0111] The optimal value is the individual with the highest current fitness.

[0112] Step 6: Generate a random perturbation (using a random function) in the neighborhood of the optimal value obtained from the new generation population in Step 5 to obtain an updated solution; determine the energy change between the updated solution and the optimal solution given by the genetic algorithm.

[0113] If the energy of the updated solution is less than the energy of the previously given optimal solution, then the updated solution replaces the optimal solution, and the latest solution is obtained.

[0114] If the energy of the updated solution is greater than or equal to the energy of the previously given optimal solution, then proceed according to probability. The updated solution is retained, and the retained updated solution is the current latest solution;

[0115] In the formula The value of the change in energy. The current temperature;

[0116] Step 7: Determine if the latest solution meets the convergence condition of the simulated annealing algorithm. If not, repeat step 6. If it does, return the latest solution to step 3 to replace the individual with the highest fitness in step 3 (initially, a certain number of solutions were randomly selected from a certain range, and after performing the selection, crossover, and mutation operations, a bunch of new solutions were generated, and we found the optimal solution. If the iteration requirement is not met, this latest solution and the previously generated bunch of solutions are used as the population again, and the crossover and mutation operations are performed, which is step 4). Repeat steps 4 to 7 until the convergence condition of the genetic algorithm is met.

[0117] Specific Implementation Method Two: This implementation method differs from Specific Implementation Method One in that the adaptive crossover in step one is:

[0118]

[0119] In the formula For adaptive coefficients, This represents the current average fitness value of the population. This represents the current maximum fitness value for the race. This corresponds to the current situation of population dispersion.

[0120] The other steps and parameters are the same as in Specific Implementation Method 1.

[0121] Specific Implementation Method Three: This implementation method differs from Specific Implementation Method One or Two in that the mutation operator in step one is:

[0122]

[0123] In the formula These are adaptive coefficients.

[0124] Other steps and parameters are the same as in specific implementation method one or two.

[0125] Specific Implementation Method Four: This implementation method differs from Specific Implementation Methods One to Three in that the initial temperature in step one is:

[0126] Based on the required solution accuracy, if further increasing the initial temperature does not significantly improve the solution accuracy, then the lower temperature at which the required accuracy is first achieved is selected as the initial temperature.

[0127] The other steps and parameters are the same as those in one of the specific implementation methods one to three.

[0128] Specific Implementation Method Five: This implementation method differs from one of Specific Implementation Methods One to Four in that the cooling coefficient in step one is:

[0129] A cooling coefficient is used to achieve a proportional decrease in temperature, and the rate of temperature decrease is controlled by the cooling coefficient.

[0130] The adaptive cooling acceleration coefficient is shown in the following formula:

[0131]

[0132] In the formula For adaptive cooling acceleration coefficient, The current temperature. This is the cutoff temperature;

[0133] Substitute into the cooling function In this process, we obtain the adaptive cooling function:

[0134]

[0135] In the formula This is the gain coefficient. Let be the temperature at time i.

[0136] The other steps and parameters are the same as those in one of the specific implementation methods one to four.

[0137] Specific Implementation Method Six: This implementation method differs from Specific Implementation Methods One through Five in that: the range of the Markov chain length in step one is... .

[0138] The other steps and parameters are the same as those in one of the specific implementation methods one to five.

[0139] Specific Implementation Method Seven: This implementation method differs from one of the specific implementation methods one to six in that the cutoff temperature in step one is 0.001.

[0140] The other steps and parameters are the same as those in one of the specific implementation methods one to six.

[0141] Specific Implementation Method Eight: This implementation method differs from one of the specific implementation methods one to seven in that the encoding method in step two is binary encoding, floating-point encoding, or symbolic encoding.

[0142] The other steps and parameters are the same as those in any of the specific implementation methods one to seven.

[0143] Specific Implementation Method Nine: This implementation method differs from Specific Implementation Methods One to Eight in that the rule selected in step four is roulette wheel rule, random competition rule, best retention method, or exclusion selection, etc.

[0144] The intersection methods include single-point intersection, two-point intersection, multi-point intersection, or uniform intersection, etc.

[0145] The mutation methods include basic bit mutation or boundary mutation, etc.

[0146] The other steps and parameters are the same as those in one of the specific implementation methods one to eight.

[0147] Simulation test

[0148] The model is based on the actual size of a wind farm. Due to the scanning characteristics of multibeam imaging sonar, the USV must move in a circle around each wind turbine. Therefore, the problem is transformed into an algorithm that needs to select five points on a circle formed by five wind turbines based on the multibeam scanning distance, and minimize the total distance of the line connecting them. This can be understood as a TSP problem involving permutations and combinations.

[0149] This paper selects 20 equally spaced points on each path circle as candidate feasible solutions. Therefore, the number of solutions to the problem is... This refers to the population size. This is a relatively large number, so feasible solutions are first screened. Since the danger zone is an inaccessible space, we first calculate whether each alternative path will enter the danger zone, and whether the distance from the alternative path's path to the center of the five wind turbine piles is less than the radius of the danger zone. Then, we update the alternative paths and the population size. Coefficients in the adaptive crossover operator Coefficients in adaptive genetic operators initial temperature Basic cooling coefficient Markov chain length Cut-off temperature To verify the effectiveness and superiority of the improved simulated annealing algorithm, simulation experiments were first conducted using genetic algorithm and simulated annealing algorithm respectively, and then simulation experiments were conducted using the improved genetic simulated annealing algorithm.

[0150] Simulation results using genetic algorithms and simulated annealing algorithms are as follows: Figure 7a , 7b ;

[0151] Then, the simulation results were obtained using the improved genetic simulated annealing algorithm proposed in this paper, as shown in Figure 8.

[0152]

[0153] The table above shows the results obtained by the three algorithms. It can be seen that the path length obtained by the traditional genetic algorithm is 16.95, and the path length obtained by the traditional simulated annealing algorithm is 17.37. The genetic algorithm's result is slightly better than the simulated annealing algorithm, which is due to the genetic algorithm's strong global space search capability, but the optimization margin is only 2%. The improved genetic simulated annealing algorithm, however, yields a path length of 15.51, representing an optimization margin of 8% compared to the genetic algorithm. This demonstrates that the improved GA-SA algorithm proposed in this invention can significantly improve the algorithm's accuracy compared to the traditional algorithm. Based on the above analysis, the improved genetic simulated annealing algorithm effectively complements the advantages and disadvantages of the two algorithms, ultimately improving the algorithm's accuracy.

[0154] This invention may have other embodiments. Without departing from the spirit and essence of this invention, those skilled in the art can make various corresponding changes and modifications according to this invention, but these corresponding changes and modifications should all fall within the protection scope of the appended claims.

Claims

1. An improved GA-SA path planning method for unmanned surface vessels used for offshore wind farm inspection, characterized in that: Step 1: Initialize the parameters of the improved genetic simulation annealing algorithm for the unmanned surface vessel used for offshore wind farm inspection, including population size, adaptive crossover operator, adaptive mutation operator, initial temperature, adaptive cooling coefficient, Markov chain length, and cutoff temperature. The population is a collection of paths taken by unmanned surface vessels. The population size is the total number of unmanned surface vessel paths; Step 2: Based on the characteristics of the improved GA-SA path planning problem solution of the unmanned surface vessel for offshore wind farm inspection, perform chromosome encoding operation and set the fitness function; The fitness function is the formula for calculating the path length of the unmanned surface vessel. Step 3: Decode all chromosomes and calculate the fitness value of all individuals, then select the individual with the highest current fitness as the optimal individual; The individual refers to the path of the unmanned surface vessel; Step 4: Determine whether the current population meets the convergence condition. If it does, the algorithm ends and outputs the current best individual. If the conditions are not met, the process enters the evolutionary stage, where selection, crossover, and mutation operations are performed. Step 5: Use the optimal value obtained from the evolved new generation population as the initial solution for the simulated annealing algorithm, and then execute Step 6; The optimal value is the individual with the highest current fitness. Step 6: Generate a random perturbation in the neighborhood of the optimal value obtained from the new generation population in Step 5 to obtain an updated solution; determine the energy change between the updated solution and the optimal solution. If the energy of the updated solution is less than the energy of the optimal solution, then the updated solution replaces the optimal solution, and the latest solution is obtained. If the energy of the updated solution is greater than or equal to the energy of the optimal solution, then proceed according to probability. The updated solution is retained, and the retained updated solution is the current latest solution; In the formula The value of the change in energy. The current temperature; Step 7: Determine whether the latest solution meets the convergence condition of the simulated annealing algorithm. If not, repeat step 6. If it does, return the latest solution to step 3 to replace the individual with the highest fitness in step 3, and repeat steps 4 to 7 until the convergence condition of the genetic algorithm is met. The adaptive crossover operator in step one is: In the formula For adaptive coefficients, This represents the current average fitness value of the population. This represents the current maximum fitness value for the race. This corresponds to the current situation of population dispersion; The adaptive mutation operator in step one is: In the formula These are adaptive coefficients; The adaptive cooling coefficient in step one is: A cooling coefficient is used to achieve a proportional decrease in temperature, and the rate of temperature decrease is controlled by the cooling coefficient. The adaptive cooling acceleration coefficient is shown in the following formula: In the formula For adaptive cooling acceleration coefficient, The current temperature. This is the cutoff temperature; Substitute into the cooling function In the process, the adaptive cooling coefficient is obtained: In the formula This is the gain coefficient. Let be the temperature at time i.

2. The improved GA-SA path planning method for unmanned surface vessels used for offshore wind farm inspection according to claim 1, characterized in that: The initial temperature in step one is: Based on the required solution accuracy, if the accuracy of the solution does not improve when the initial temperature is further increased, the lower temperature at which the accuracy is first achieved is selected as the initial temperature.

3. The improved GA-SA path planning method for unmanned surface vessels used for offshore wind farm inspection according to claim 2, characterized in that: The range of values ​​for the Markov chain length in step one is: .

4. The improved GA-SA path planning method for unmanned surface vessels used for offshore wind farm inspection according to claim 3, characterized in that: The cutoff temperature in step one is 0.

001.

5. The improved GA-SA path planning method for unmanned surface vessels used for offshore wind farm inspection according to claim 4, characterized in that: The encoding method in step two is binary encoding, floating-point encoding, or symbolic encoding.

6. The improved GA-SA path planning method for unmanned surface vessels used for offshore wind farm inspection according to claim 5, characterized in that: In step four, the rule to be selected is roulette wheel rule, random competition rule, best retention method, or exclusion selection; The crossing methods include single-point crossing, two-point crossing, multi-point crossing, or uniform crossing; The mutation methods are basic bit mutation or boundary mutation.

Images