摘要
某型固定翼无人机(UAV)纵向控制回路存在典型的多工况特征。针对单一模型对飞机高度值进行预测会导致模型失配,准确度不高的问题,提出了“聚类分析-模式识别-回归预测”相结合的多工况状态预测技术路线。该技术以俯仰角、升降舵偏角、缸温、发动机转速为预测变量,通过离线建模与在线预测两个阶段完成飞机高度预测。离线建模阶段,使用基于共享近邻相似性的密度聚类算法将UAV训练数据分为多个工况,之后使用核高斯过程回归(KGPR)算法建立各工况预测模型。在线预测阶段,使用人工神经网络算法完成测试数据工况辨识,并调用相应的KGPR模型完成状态预测。仿真试验选取了代表UAV典型用途的近程、中程、远程三种实飞数据,并以此为基础对算法进行了验证,结果表明所提技术能够有效提高预测准确度,具有实用价值。
A certain type of fixed-wing UAV has typical multi-working condition characteristics.It will result in the model mismatch when a single model is used to predict the aircraft height value.Aiming at this problem,a new technical scheme of UAV multi-working condition prediction based on“cluster analysis-pattern recognition-regression prediction”was proposed.The novel technical scheme takes pitch angle,elevator yaw angle,engine cylinder temperature,and engine speed as predictive variables and completes the altitude prediction through two stages of offline modeling and online prediction.In the offline modeling stage,the shared nearest neighbor based density clustering algorithm was used to analyze the UAV training data and to separate the flight data into multiple working conditions.Then the nuclear Gaussian process regression(KGPR)algorithm was used to establish prediction model of each working condition.In the online prediction stage,the artificial neural network algorithm was used to complete the test data working condition identification,and the corresponding KGPR model was invoked for state prediction.Three kings of real flight representing the typical uses of UAV were selected in the simulation experiments to verify the algorithms.The results show that the proposed technique scheme can effectively improve the prediction accuracy and has practical value.
作者
梁少军
郑幸
谢礼鹏
林冬生
LIANG Shaojun;ZHENG Xing;XIE Lipeng;LIN Dongsheng(School of Ordnance Sergeant Army Engineering University,Wuhan 430075,China;School of Electrical Engineering and Automation,Wuhan University,Wuhan 430072,China)
出处
《兵器装备工程学报》
CAS
北大核心
2020年第10期203-209,共7页
Journal of Ordnance Equipment Engineering
关键词
多工况
固定翼无人机
密度聚类
回归预测
人工神经网络
multiple condition
fixed-wing uav
DBSCAN
regression prediction
artificial neural network
