Intelligent Scheduling Technology of Swarm Intelligence Algorithm for Drone Path Planning
[ad_1]
2.1. Theory of Selected Algorithms
Mimicking the swarm intelligence behavior of ants in determining the closest route to target food, the ACO algorithm developed by Dorigo is the earliest SI algorithm with practical significance.
where denotes pheromone intensity at time and refers to the pheromone volatilization factor.
Inspired by the foraging behavior of birds, the PSO algorithm proposed by Eberhart is one of the most widely used SI algorithms.
Here, is the inertia weight, and are the learning rates, the historical optimal position of the corresponding particle, and the global optimal position in this iteration. As a coefficient, determines the influence of the particle velocity of the previous iteration for this iteration.
Classical algorithms, including ACO and PSO, have a strong robustness and global search ability, but also a slow convergence speed and poor accuracy. From the perspective of model structure, the slow convergence speed is due to the single structure. There is no hierarchical stratification of information among individuals. Although the equivalence of information can enhance the search ability of the algorithm in the early stage, once the local optimum is found in the later stage, it is easy to break the balance between individuals and affect the adjacent individuals, so that the algorithm falls into the local optimum. Although the accuracy is low, due to a less random mechanism, it has a shorter computing time in the low-dimension route planning optimization problem.
Newer algorithms, such as GWO and WOA, whose superiority is attributed to the more diverse and efficient algorithm structure design, have fast convergence speeds and high accuracies and have been widely used in flight-path planning.
Here, , , and denote the position vectors of wolves , , and δ in the current population; is the location of individual grey wolves; , , and denote the distances between the current object wolf and the three candidate wolves; and are coefficient vectors; refers to the position vector of the prey; is the distance between the current grey wolf and the location of the prey.
Here, denotes the number of iterations, and the updated coefficient vectors, the optimal position of the whale at iteration , the position vector of the current whale. The first two formulas show the shrinkable encircling prey stage, and the final formula is the spiral position update.
WOA is one SI algorithm with many random search mechanisms, and the selection of the information updating formula is random. Many random search mechanisms can effectively improve the global search ability in the early stage of the algorithm, but simultaneously affect the stability of the algorithm performance and computation time in the late stages.
These four algorithms cover the early SI algorithms that have been widely used and new SI algorithms proposed in recent years. From the aspect of algorithm performance, they cover different performance requirements, such as strong robustness, strong convergence, and high accuracy. Most importantly, they are widely used in the field of drone cluster control. In summary, the four algorithms were selected as candidate algorithms and incorporated into the algorithm scheduling library for the DQN to study algorithm scheduling technology.
[ad_2]