An unsupervised learning neural network for planning UAV full-area reconnaissance path

一种无监督学习型神经网络的无人机全区域侦察路径规划

¹ School of Electronics and Information, Northwestern Polytechnical University, Xi'an 710129, China
² CETC Key Laboratory of Data Link Technology, Xi'an 710068, China
³ AVIC Xi'an Aeronautics Computing Technique Research Institute, Xi'an 710068, China

Received: 15 June 2020

Abstract

To plan a UAV's full-area reconnaissance path under uncertain information conditions, an unsupervised learning neural network based on the genetic algorithm is proposed. Firstly, the environment model, the UAV model and evaluation indexes are presented, and the neural network model for planning the UAV's full-area reconnaissance path is established. Because it is difficult to obtain the training samples for planning the UAV's full-area reconnaissance path, the genetic algorithm is used to optimize the unsupervised learning neural network parameters. Compared with the traditional methods, the evaluation indexes constructed in this paper do not need to specify UAV maneuver rules. The offline learning method proposed in the paper has excellent transfer performances. The simulation results show that the UAV based on the unsupervised learning neural network can plan effective full-area reconnaissance paths in the unknown environments and complete full-area reconnaissance missions.

摘要

针对不确定信息条件下的无人机全区域侦察路径规划问题，提出了一种基于遗传算法的神经网络无监督学习方法。构建了环境模型、无人机模型与评价指标，并建立了无人机全区域侦察路径规划的神经网络模型。在此基础上，考虑到路径规划神经网络训练样本难以获取的问题，采用遗传算法对神经网络参数的无监督训练进行了优化。相对于传统方法，文中构建的评价指标不需要人为指定无人机机动规则，并且提出的离线学习方法具有优良的迁移能力。仿真结果表明，基于无监督学习型神经网络的无人机，能够在未知环境中规划出有效路径完成全区域侦察任务。