基于力系简化神经网络的过驱动控制分配  

Over-actuated System Control Allocation Based on Simplified Force System Neural Network

作　　者：任军洲 蔡勇[1,2] 李自胜[1] REN Junzhou;CAI Yong;LI Zisheng(School of Manufacturing Science and Engineering,Southwest University of Science and Technology,Mianyang 621000,China;Key Laboratory of Testing Technology for Manufacturing Process of Ministry of Education,Mianyang 621000,China)

机构地区：[1]西南科技大学制造科学与工程学院,绵阳621000 [2]制造过程测试技术省部共建教育部重点实验室,绵阳621000

出　　处：《自动化与仪表》2023年第9期41-45,50,共6页Automation & Instrumentation

基　　金：国家重点研发计划项目(2021YFB3400702);四川省科技计划项目(2018GZ0083,2018JY0245)。

摘　　要：针对过驱动系统的控制分配问题,该文提出直接利用神经网络实现最优控制分配。首先,结合空间力系简化原理以及自编码器训练方法,提出力系简化神经网络结构;然后,利用双循环训练流程和均匀分布随机生成数据集训练神经网络;最后,基于单刚体控制分配任务验证方法可行性。在实验部分,总结出适用于控制分配任务的损失函数和学习率更新方法,并得到神经网络模型。悬浮控制实验和轨迹跟踪控制实验证明训练得到的神经网络模型可用于过驱动系统的控制分配,并且相比二次规划,不存在偏见,同时能减少40%多余控制量。In order to solve the control allocation problem of the over-actuated system.It is proposed to directly use neural networks to achieve optimal control allocation.Firstly,combined with the principle of spatial force system simplification and the auto-encoder training method,the structure of the simplified neural network with force system is proposed.Then,train the neural network using the double-loop training process and evenly distributed randomly generated dataset.Finally,the feasibility of the method is verified based on the single rigid body control allocation task.In the experimental part,the loss function and learning rate update methods suitable for controlling the allocation tasks are summarized,and the neural network model is obtained.The suspension control experiment and the trajectory tracking control experiment show that the trained neural network model can be used for the control allocation of the over-actuated system,and compared with the quadratic programming,there is no bias,and the excess control amount can be reduced by 40%.

关 键 词：力系简化 神经网络 过驱动系统 控制分配 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]