基于强化学习的飞航导弹姿态控制PID参数调节方法

PID Attitude Controller Tuning for Aerodynamic Missile Based on Reinforcement Learning

针对人工智能在传统飞航导弹控制系统优化改造方面的应用进行了初步探索。考虑到目前实际应用中控制结构基本上都是采用PID控制形式,利用强化学习技术实现了PID姿态控制参数的在线闭环快速自适应整定。通过分别搭建评价网络和动作执行网络,实现对非线性系统最优目标函数的逼近和对PID控制参数的拟合,并利用梯度下降法给出了不基于被控对象动力学而只依赖于观测数据的神经网络权值调整公式。考虑到实际工程实现,动作执行网络的输入采用长周期信号,数学仿真结果验证了所提算法的有效性。

Preliminary investigation on traditional control system design upgrading of aerodynamic missile based on artificial intelligence is performed.In practice,the most popularly used control structure is PID controller.Therefore,the reinforcement learning technique is utilized to realize the fast tuning of PID attitude controller online in an adaptive manner.The critic network and the action network are established to approximate the infinite horizon cost function and interpolate PID controller parameters,respectively.The weights of both networks are modified by only using the measurement information without any knowledge about the flight dynamics according to the gradient descent method.Taking the practical implementation into account,the inputs of the action network are selected as the long-period variables.Simulation results demonstrate the effectiveness of the proposed method.