基于异步优势执行器评价器的自适应PID控制

The adaptive PID controller based on Asynchronous Advantage Actor-Critic learning

摘要:

自适应PID较好地解决了传统PID无法自整定参数的问题,已成为控制领域内的研究热点。研究基于异步优势执行器评价器(Asynchronous Advantage Actor-Critic,A3C)算法设计了一种新的自适应PID控制器。该控制器利用A3C结构的多线程异步学习特性,并行训练多个执行器评价器(Actor-Critic,AC)结构的智能体,每个智能体采用多层前馈神经网络逼近策略函数和值函数实现在连续动作空间中搜索最优的参数整定策略,以达到最佳的控制效果。与已有的多种自适应PID控制器性能对比分析结果表明该方法具有收敛速度快,自适应能力强的特点。

Abstract:

Self-adaptive PID has become a hotspot in the field of control, it can solve the problem that traditional PID can’t turning parameters. This paper proposed a new adaptive PID controller based on the Asynchronous Advantage Actor-Critic (A3C) algorithm. It used the multi-threaded and asynchronous learning style to train multiple agents of Actor-Critic (AC) structures in parallel. In order to achieve the best effect, each agent adapts a multilayer feedforward neural network to approximate strategy function and value function. In this way, they can search for the best parameter turning strategies in continuous motion space. Compared with the performance of others adaptive PID controllers, the results show that this method has the advantage of fast convergence and strong self-adaptability.