Intelligent vehicle path planning method based on improved whale optimization algorithm
The improved whale optimization algorithm addresses issues of route diversity and local optima in autonomous vehicle path planning by using Levy flight initialization and Brownian motion perturbation, enhancing accuracy and efficiency while reducing hardware consumption.
Patent Information
- Application Number
- CN202510582798.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-05-07
- Publication Date
- 2025-07-15
AI Technical Summary
The existing whale optimization algorithms have problems such as insufficient diversity of initial route selection, easy to fall into local optimality and low efficiency in finding optimal routes in the path planning of autonomous vehicles, resulting in high redundancy of planned routes, increased hardware energy consumption and difficult to meet the needs in real time.
The Levi flight strategy is used to initialize the position of the intelligent vehicle, combine Brownian motion to randomly perturb the iteration process, and build and improve whale optimization algorithm to enrich the diversity of initial information and avoid local optimal traps, and improve global search capabilities.
It improves the accuracy and efficiency of path planning of autonomous driving vehicles, reduces hardware resource consumption, meets real-time requirements, and reduces path redundancy.
Smart Images
Figure CN120313630A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of intelligent transportation and relates to an intelligent vehicle path planning method based on an improved whale optimization algorithm. The method is applicable to vehicle navigation systems and autonomous driving technologies and is used to improve global search accuracy and reduce hardware resource consumption. Background Art
[0002] Autonomous vehicles have important application value in modern transportation systems, and can improve traffic safety, alleviate traffic congestion, and improve travel efficiency. However, autonomous vehicles face many challenges in actual operation. For example, in complex urban traffic environments, vehicles need to accurately perceive the surrounding environment and plan safe and efficient driving routes. However, there are many problems in actual traffic scenarios, such as dynamic changes in traffic flow, road construction, and sudden accidents. These problems restrict the perception accuracy of autonomous vehicles and the accuracy of route planning.
[0003] At present, for the perception and planning problems of autonomous vehicles in complex environments, swarm intelligence optimization methods use multiple entities to collaboratively search for the optimal route and interact with information in the solution space, and finally achieve the global optimization goal. Compared with traditional gradient optimization methods or evolutionary methods, swarm intelligence optimization methods have the advantages of simple parameter configuration and low computing resource usage in embedded hardware implementation. Classic swarm intelligence optimization methods (such as ant colony optimization, particle swarm optimization, artificial bee colony, etc.) have been applied to the field of industrial equipment control, and whale optimization (WOA), as an emerging method, has been integrated into hardware methods such as vehicle route planning modules and reservoir scheduling controllers due to its few parameters and strong optimization ability. However, the existing hardware implementation methods based on WOA have significant defects: first, the initialization of intelligent vehicles is simplistic. Traditional WOA uses fixed random distribution to initialize intelligent vehicles in embedded devices, resulting in insufficient diversity in initial route selection. Secondly, there is the local optimal route trap. This method lacks a hardware-friendly perturbation route mechanism, and the path planning results are prone to fall into local optimality. In complex terrain scenes, the average redundancy of the planned path length is 18%, and the real-time performance is difficult to meet the millisecond response requirements. Finally, the efficiency of finding the optimal route is inefficient. In programmable logic devices such as FPGA, traditional WOA requires more than 80 rounds of convergence due to the rigid iterative logic, resulting in a 25%-40% increase in hardware energy consumption.
[0004] In view of the above problems, the present invention improves it based on Levy flight and Brownian motion, and proposes an intelligent vehicle path planning method based on an improved whale optimization algorithm. On the one hand, the Levy flight method is used to optimize the initialization process of intelligent vehicles, and by enriching the diversity of the initial information of intelligent vehicles, the accuracy of solving the optimal route and the efficiency of finding the optimal route of the traditional WOA are improved; on the other hand, the principle of Brownian motion is used to randomly perturb the position of the intelligent vehicle after updating during the iteration process, so as to prevent the intelligent vehicle route from falling into the local optimum. Summary of the Invention
[0005] In view of this, the technical problem to be solved by the present invention is to propose an intelligent vehicle path planning method based on an improved whale optimization algorithm, which can solve the problems of low accuracy in solving the optimal route, slow efficiency in finding the optimal route, and easy to fall into the local optimal route in the route planning of intelligent vehicles.
[0006] The technical solutions adopted by the present invention to solve the technical problems are as follows:
[0007] An intelligent vehicle path planning method based on an improved whale optimization algorithm, which includes:
[0008] Step 1) Construct an initialization module based on the Levy flight strategy, simulate the Levy flight motion according to the Mantegna method, and randomly generate the positions of intelligent vehicle individuals, so as to realize the initialization of the position information of intelligent vehicles and the surrounding environment information;
[0009] Step 2) Construct a route selection and update module. In each iteration, evaluate the fitness value of the intelligent vehicle individuals to update the optimal route, check the number of iterations to decide whether to continue, and calculate parameters based on the current optimal route to determine the subsequent route update strategy;
[0010] Step 3) Construct a perturbation module combined with the principle of Brownian motion, and randomly perturb the position update of the intelligent vehicle during the iteration based on Brownian motion.
[0011] Further, the specific steps of the step 1) include the following steps:
[0012] Step 1.1) Randomly generate the first individual X0. This individual is used as the starting point of the intelligent vehicle, and its position information and surrounding environment information are randomly determined.
[0013] Step 1.2) The initialization strategy of the intelligent vehicle based on Levy flight. For each subsequent individual X in the intelligent vehicle i+1 , its position is based on the previous individual X iThe position is determined by Lévy flight motion. Lévy flight is a random walk process, characterized by step lengths following the Lévy distribution, which has heavy-tailed characteristics, i.e., very large step lengths occasionally occur, which helps the intelligent vehicle to explore new areas in the search space. Directly using the Lévy distribution to generate step lengths is complex because the mathematical expression of the Lévy distribution is complex and difficult to sample directly. To simplify the implementation of Lévy flight, the Mantegna method is commonly used to simulate the Lévy flight process. In the Mantegna method, the position of individual X i+1 is based on the position of X i , plus an offset determined by u and v. Among them, u and v respectively follow the normal distribution with parameters and σ u and σ v . The calculation formula gives how to determine the parameters of these normal distributions according to the characteristics of Lévy flight. In this method, β is set to the empirical value 1.5, and this value can provide good search performance.
[0014] Furthermore, step 2) specifically includes the following steps:
[0015] Step 2.1) Evaluate the fitness value and update the optimal route. At the beginning of each iteration, this module first calculates the fitness value of each individual in the current intelligent vehicle one by one. These fitness values are calculated through the fitness function and are used to measure the superiority and inferiority of individuals in a given problem. Based on these calculated fitness values, this module updates the optimal route in the current iteration, that is, selects the individual with the best performance in the current intelligent vehicle as the current optimal route.
[0016] Step 2.2) Iteration control. Next, this module checks whether the current iteration round is less than the preset maximum number of iterations. This is a key step in iteration control and is used to decide whether to continue the subsequent iteration process. If the current iteration round is less than the preset maximum number of iterations, then this module will continue to execute the subsequent steps; otherwise, this module will stop the iteration and output the currently found optimal route as the final result.
[0017] Step 2.3) Calculate parameters to determine the update strategy. In each iteration, in addition to evaluating the fitness value and performing iteration control, this module also calculates parameter A based on the current optimal route and generates a random number p. These parameters (including the coefficient vector) will jointly determine the subsequent update strategy in order to expect to find a better route in the subsequent iterations.
[0018] Furthermore, step 3) specifically includes the following steps:
[0019] Construct a perturbation module that combines the principle of Brownian motion. This module randomly perturbs the positions of the optimal and sub-optimal intelligent vehicles in the current iteration by introducing Brownian motion, thereby replacing the positions of the worst and second-worst intelligent vehicles, avoiding the route from falling into local optimality, and enhancing the global search ability. Specifically, the calculation of the position and perturbation is based on the Brownian motion expression, and the optimal and sub-optimal intelligent vehicles are determined by sorting in descending order using sort, with the relevant parameters set as fixed values. Brief Description of the Drawings
[0020] In order to make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings, where:
[0021] Figure 1 is the flowchart of this method;
[0022] Figure 2 is the overall model framework diagram of this method;
[0023] Figure 3 is the Lévy flight initialization module constructed by this method;
[0024] Figure 4 is the vehicle position update module with a spiral update method constructed by this method;
[0025] Figure 5 is the vehicle position update module with a contraction and enclosure method constructed by this method;
[0026] Figure 6 is the vehicle position update module with a search and foraging method constructed by this method;
[0027] Figure 7 is the Brownian motion perturbation module constructed by this method;
[0028] Figure 8 is the convergence curve graph of WOA, PSO, GA, ACO, and LBWOA under the benchmark test function F1. Specific Embodiments
[0029] The preferred embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings.
[0030] An intelligent vehicle path planning method based on an improved whale optimization algorithm provided by the present invention. The flowchart and overall model block diagram of this method are as shown in Figure 1 and Figure 2 shown, and this method includes the following steps:
[0031] Step 1) Hardware configuration and initialization settings of parameters;
[0032] Step 2) Initialize the intelligent vehicle information using the Lévy flight initialization module;
[0033] Step 3) According to the set round n t Loop and update the optimal route;
[0034] Step 4) Use modules such as search foraging, shrinking enclosing, and spiral update to update the position of the intelligent vehicle;
[0035] Step 5) Use the Brownian motion perturbation module to randomly perturb the position of the intelligent vehicle;
[0036] Step 6) Select methods such as WOA, PSOA, GA, and ACOA as comparison objects of LBWOA, and output the optimal route.
[0037] Furthermore, the specific content of Step 1) includes the following steps:
[0038] Step 1.1) The platform hardware is configured with a 2.2GHz quad-core Intel Core i7 processor and 16GB 1600MHz DDR3 memory, and is implemented by programming with Python 3.7.7.
[0039] Step 1.2) Set the number of intelligent vehicles to 30 and the maximum number of iterations to 100. To reduce the influence of errors on the contingency of the results, each group runs independently 50 times.
[0040] Furthermore, the specific content of Step 2) includes the following steps:
[0041] Step 2.1) Randomly generate the first random intelligent vehicle X0 and the surrounding environment information.
[0042] Step 2.2) As Figure 3 shown, the position of each intelligent vehicle individual X i+1 is based on the position of the previous individual X i to perform Levy flight motion, as shown in formula (1):
[0043]
[0044] where: l is the step size control quantity; Levy(λ) is the random search path, which satisfies formula (2):
[0045] Levy(λ)~u=i -λ 1<λ≤3 (2)
[0046] Since the Levy flight described by formulas (1) and (2) is too complex to be applied and implemented in real scenarios, the Mantegna method is commonly used to simulate the Levy flight process. The position calculation process of X i+1 is shown in formula (3):
[0047]
[0048] where u and v respectively follow normal distributions with parameters σ u and σ v . The definitions of σ u and σ v are shown in formula (4):
[0049]
[0050] In this method, β in formula (4) is set to the empirical value 1.5.
[0051] Furthermore, the specific content of step 3) includes the following steps:
[0052] Step 3.1) Calculate the fitness value of each individual of the intelligent vehicle one by one: At the beginning of each iteration, this module will evaluate the fitness value of each individual in the current intelligent vehicle. These values are calculated based on the fitness function and are used to measure the quality of the individual in a given problem. Based on the calculated fitness values, this module will update the current optimal route, which is the optimal route in the current iteration.
[0053] Step 3.2) Check whether the current round i is less than n t , which is a key step in iterative control. If the current round i is less than n t , then this module will continue to execute the subsequent steps; otherwise, this module will stop the iteration and output the optimal route.
[0054] Step 3.3) In each iteration, this module will calculate parameter A and generate a random number p based on the current optimal route. A is a coefficient vector, and its calculation process is shown in formula (5):
[0055]
[0056] These parameters will be used to determine the subsequent optimal route update strategy.
[0057] Furthermore, the specific content of step 4) includes the following steps:
[0058] Step 4.1) As Figure 4 shown, when p < 0.5, spiral update is adopted. The spiral update method mainly performs spiral motion on the optimal intelligent vehicle individual to refine the search for the optimal route. In this method, the position update process of the intelligent vehicle individual is as shown in formula (6):
[0059]
[0060] where the parameter b is the constant coefficient of the logarithmic spiral, l is a random number between [-1, 1], D eRepresents the determined distance between the intelligent vehicle individual and the optimal individual.
[0061] Step 4.2) As Figure 5 shown, when p≥0.5 and |A|<1, shrinkage enclosure is adopted. In the shrinkage enclosure mode, the intelligent vehicle individual updates its position in the next round based on the position of the optimal intelligent vehicle in the previous round and actively approaches the optimal position. The specific position update is shown in formula (7):
[0062]
[0063] Where is the position of the best intelligent vehicle m in the t-th round.
[0064] Step 4.3) As Figure 6 shown, when p≥0.5 and |A|≥1, search foraging is adopted. In the t-th iteration, the position of the i-th best intelligent vehicle is expressed as and is updated according to the position of a randomly selected other intelligent vehicle r. In the (t + 1)-th iteration of this stage, the position of the i-th best intelligent vehicle can be calculated by formula (8):
[0065]
[0066] Where: is the position of the random intelligent vehicle r in the t-th iteration; A is the coefficient vector: D r is the random distance between intelligent vehicle i and intelligent vehicle r in the current round, and its calculation process is shown in formula (9):
[0067]
[0068] Here: r1 and r2 are random vectors, and the component value range is [0, 1]; E represents the identity matrix; t max is the maximum number of iterations; " " represents an operation method, such as
[0069] It can be seen from formula (8) that as the iteration round t increases, the value of parameter a linearly decreases from 2 to 0. Correspondingly, the expected value of the absolute value of the coefficient vector A also linearly decreases from 2 to 0. In formula (9), the component of C takes a random number between [0, 2] to control the influence of the random intelligent vehicle individual on the current individual position.
[0070] Furthermore, the specific content of step 5) includes the following steps:
[0071] As Figure 7As shown, this method (LBWOA) introduces Brownian motion to randomly perturb the updated positions of intelligent vehicles, so as to prevent this method from prematurely falling into local optimal routes, thereby enhancing the ability of this method to explore global optimal routes. The specific method is to calculate the positions of the optimal intelligent vehicle and the sub-optimal intelligent vehicle in the current iteration round according to formulas (10) and (11), and replace the positions of the worst intelligent vehicle and the sub-worst intelligent vehicle with the calculation results.
[0072]
[0073]
[0074] Among them, the sort function represents descending sorting of the set. Assume that the number of intelligent vehicles in the population is n, and after sorting, and Y1 k are the optimal intelligent vehicle and the sub-optimal intelligent vehicle respectively, and are the worst intelligent vehicle and the sub-worst intelligent vehicle respectively. In this method, σ and t are set to 0.25 and 10 respectively.
[0075] Furthermore, the specific content of step 6) includes the following steps:
[0076] A comparative experiment is conducted on LBWOA and four methods, namely WOA, PSO, GA, and ACO. The convergence curves obtained by running the five methods on the benchmark test function F1 are as Figure 8 shown.
[0077] Analysis Figure 8 shows that on F1, LBWOA can obtain a better route than methods such as WOA, PSO, GA, and ACO. LBWOA rapidly achieved an accuracy improvement of multiple orders of magnitude within 100 rounds, which indicates that LBWOA is faster in finding the optimal route.
[0078] Finally, it should be pointed out that the above implementation cases are only used to illustrate the technical solutions of the present invention and are not intended to impose any restrictions on it. Although the present invention has been comprehensively and carefully described through the foregoing implementation cases, professionals in the field should be aware that various adjustments can be made in its manifestations and specific details, and these adjustments will not exceed the scope defined by the claims of the present invention.
Claims
1. An intelligent vehicle path planning method based on an improved whale optimization algorithm, characterized in that, Including the following steps: Step 1: This method is implemented on a hardware platform with a 2.2GHz quad-core Intel Core i7 processor and 16GB 1600MHz DDR3 memory, using Python 3.7.7 as the programming language to implement the relevant method. First, set parameters to ensure the accuracy and stability of the results: the number of intelligent vehicles is set to 30, which means that in each iteration, this method will simultaneously process 30 potential route solutions; at the same time, the maximum number of iterations is limited to 100 times to ensure that the method can converge to the optimal route within a reasonable time. To minimize the impact of random errors on the final results, this method adopts a strategy of repeated experiments, that is, each group of experiments runs independently 50 times, and statistical analysis is performed on these running results to obtain more reliable and stable conclusions. Step 2: In the initialization stage, this method first randomly generates the first individual as the starting point of the intelligent vehicle, and its position is randomly determined. Subsequently, a Levy flight initialization module is used to generate each subsequent individual in the intelligent vehicle. This module determines the position of the subsequent individual through Levy flight motion. Levy flight is a random walk process with heavy-tailed characteristics, which occasionally produces very large step sizes, helping the method to explore new regions in the search space. To simplify the implementation of Levy flight, this method uses the Mantegna method to simulate the Levy flight process, where the position of an individual is calculated based on the position of the previous individual plus an offset determined by a parameter that follows a specific normal distribution. In this method, one of the parameters of the normal distribution is set to the empirical value 1.5 to provide good search performance, thus ensuring the effective initialization of the intelligent vehicle. Step 3: At the beginning of each iteration, this method first calculates the fitness value of each individual in the current intelligent vehicle one by one. These values are obtained through the fitness function, which is used to measure the quality of an individual on a given problem, and accordingly updates the optimal route in the current iteration, that is, selects the individual with the best performance as the current optimal route. Then, this method checks whether the current iteration round is less than the preset maximum number of iterations to decide whether to continue the iteration. If the iteration condition is met, parameters are calculated based on the current optimal route and a random number is generated. These parameters (including the coefficient vector) will jointly determine the subsequent update strategy, with the expectation of finding a better route in the subsequent iteration. If the iteration condition is not met, the iteration is stopped and the currently found optimal route is output as the final result. Step 4, during the execution of the method, different update strategies are adopted according to specific conditions to find the optimal route. When p < 0.5, the spiral update method vehicle position update module proposed in this method is adopted, mainly for the optimal intelligent vehicle individual. Through the spiral update method to search for the optimal route, this module combines the constant coefficient of the logarithmic spiral, a random number in the [-1, 1] interval, and the distance between the intelligent vehicle individual and the current optimal individual, realizing a detailed search near the optimal route. When p ≥ 0.5 and |A| < 1, the contraction and enclosure method vehicle position update module proposed in this method is adopted. The intelligent vehicle individual is updated based on the position of the optimal intelligent vehicle in the previous round and actively approaches it. The search range is gradually reduced through the contraction and enclosure method, making the intelligent vehicle individual closer to the optimal route. When p ≥ 0.5 and |A| ≥ 1, the search and foraging method vehicle position update module proposed in this method is executed. Each intelligent vehicle is updated according to the position of other randomly selected intelligent vehicles. The update formula combines the position of the randomly selected intelligent vehicle in the current round, the coefficient vector, and the random distance between intelligent vehicles. This random distance is calculated by search and foraging, involving parameters such as random vectors, identity matrices, and the maximum number of iterations. Moreover, as the number of iterations increases, the expected value of the absolute value of the coefficient vector linearly decreases to 0, meaning that the movement range of the intelligent vehicle individual gradually shrinks. At the same time, by controlling the components of C (taking random numbers in the [0, 2] interval) to adjust the influence of the randomly selected intelligent vehicle individual on the current individual's position, so as to find new routes in the entire search space and maintain the diversity of routes. Step 5, the Brownian motion perturbation module proposed in this method (LBWOA) adopts a position update strategy based on Brownian motion. This module randomly perturbs the positions of the optimal and sub-optimal intelligent vehicles in the current iteration. Through this module, we replace the positions of the worst and sub-worst intelligent vehicles, thus optimizing the search process. When executing this strategy, the intelligent vehicles are first sorted in descending order by the sort function to determine the optimal and sub-optimal intelligent vehicles. At the same time, to ensure the stability and effectiveness of the strategy, relevant parameters are set to fixed values. Step 6, experiments show that LBWOA (Lévy flight and Brownian motion based whale optimization algorithm) performs better than WOA (whale optimization), PSO (particle swarm optimization), GA (genetic method), and ACO (ant colony optimization) on benchmark test functions. Specifically, LBWOA can obtain a better route and has a faster convergence speed on specific functions. Its convergence accuracy on all test functions is higher than that of the other four methods. Especially on the benchmark test functions, its convergence accuracy has increased by six to seven orders of magnitude. In addition, by introducing the Brownian motion and Lévy flight methods, LBWOA enhances the global search ability for the optimal route and the efficiency of finding the optimal route.
2. The intelligent vehicle path planning method based on the improved whale optimization algorithm according to claim 1, wherein In the Levy flight initialization module constructed in Step 2, during the iterative process of the intelligent vehicle in the optimization method, the position of each subsequent individual is determined by simulating the Levy flight movement based on the position of the previous individual. Levy flight is a random walk process with special properties, and its step size follows the Levy distribution. The characteristic of this distribution lies in its heavy tails, that is, very large step sizes will occasionally occur. This characteristic helps the method to effectively explore new potential areas in a vast search space. However, directly applying the Levy distribution to generate step sizes is quite complex because the mathematical description of the Levy distribution is rather cumbersome and it is difficult to directly sample. To simplify the implementation process of Levy flight, this method uses the Mantegna method to simulate this process. In the Mantegna method, the position of the subsequent individual is calculated based on the position of the previous individual plus an offset determined by specific parameters. This offset is composed of two random numbers that follow the normal distribution, and the parameters of these two normal distributions are determined according to the characteristics of Levy flight. This method can effectively simulate the Levy flight process, thereby utilizing this characteristic to enhance the global search ability in the optimization method.
3. The intelligent vehicle path planning method based on the improved whale optimization algorithm according to claim 1, wherein In Steps 3 and 4, the fitness values of the intelligent vehicle individuals are evaluated to update the optimal route and the parameters are calculated based on the current optimal route to determine the subsequent update module. The update module includes the vehicle position update module in the spiral update mode, the vehicle position update module in the shrinking enclosure mode, and the vehicle position update module in the search and foraging mode. In each iteration, this method first calculates the fitness values of each individual in the current intelligent vehicle one by one. These values are obtained through a specific fitness function and are used to accurately measure the performance of the individual on the given problem. Based on these fitness values, the optimal route of the current iteration is updated, and the individual with the best performance is selected. Subsequently, iterative control is performed to check whether the current iteration round has reached the preset maximum number of iterations. If not, continue the iteration and repeat the above process; if so, stop the iteration and output the currently found optimal route, which is obtained through continuous evaluation and update and represents the best solution that can be found within the current number of iterations.
4. The intelligent vehicle path planning method based on the improved whale optimization algorithm according to claim 1, wherein The Brownian motion perturbation module constructed in step 5 aims to enhance the global search ability of the method and avoid falling into local optima. Specifically, this method introduces Brownian motion to randomly perturb the positions of the optimal and sub-optimal intelligent vehicles in the current iteration. This strategy is achieved by replacing the positions of the worst and second-worst intelligent vehicles, thus promoting the diversity of intelligent vehicles. During implementation, the sort function is first used to sort the intelligent vehicles in descending order to determine the optimal and sub-optimal intelligent vehicles in the current iteration. Subsequently, the perturbation amount is calculated according to the mathematical expression of Brownian motion, which depends on the set fixed parameters to ensure the stability and repeatability of the method. By applying the calculated perturbation amount to the positions of the optimal and sub-optimal intelligent vehicles, new positions are generated, and these positions are introduced into the intelligent vehicles as candidate routes, replacing the originally worst and second-worst intelligent vehicles. This update strategy not only retains the information of excellent individuals but also introduces new search directions through random perturbation, thus effectively enhancing the global search ability of the method.