Neural Network Optimization Based On BSOWASD Algorithm And Its Application In UAV Flight Control

The core of a drone flight control system is the flight control algorithm.A welldesigned flight control algorithm enables the drone to plan a safe and reasonable flight trajectory in real-world environments,successfully avoiding obstacles and reaching the target point within an efficient timeframe.After conducting comprehensive research and study on existing flight control theories,this paper proposes a metaheuristic algorithm called Beetle Swarm Optimization(BSO)combined with the Weight and Structure Determination(WASD)algorithm.This combination is then used for training and optimization of a Feedforward Neural Network(FNN),ultimately enabling the drone to perform flight tasks in different environments and scenarios.The main contributions of this paper are as follows:(1)Traditional search algorithms often suffer from fast convergence into local optima during global search and lack the ability to escape such local optima in a timely manner,resulting in poor performance in global search.This paper introduces a metaheuristic algorithm,the BSO algorithm,which simulates the ecological behavior of beetle swarms searching for food to achieve global search.The BSO algorithm effectively avoids falling into local optima.To demonstrate the efficiency of the BSO algorithm,numerical validation experiments are conducted,showing that compared to other existing search algorithms,the BSO algorithm exhibits robustness,fast computation speed,and superior global search capability.After demonstrating the effectiveness of the BSO algorithm through numerical validation experiments,this paper combines the BSO algorithm with the WASD algorithm to construct the BSOWASD neural network training algorithm.(2)The proposed algorithm is then applied to the drone flight control module in different scenarios,and the performance and effectiveness of the algorithm in practical flight tasks are further analyzed.For single-drone flight missions,this paper constructs an FNN trained using the BSOWASD algorithm to calculate the optimal control parameters in the flight control module,enabling the drone to navigate obstacles more quickly and accurately,reaching the target point efficiently.(3)In multi-drone swarm missions,the BSOWASD algorithm is utilized to train the FNN,and the entire swarm flight mission is completed.A comparison is made with control modules that use traditional control algorithms.Experimental results from both types of missions demonstrate that the FNN trained using the BSOWASD algorithm outperforms traditional control algorithms.It achieves better control of drone flight objectives,optimizes flight trajectories,reduces errors,and accelerates the completion time of the entire flight mission.