TUNING OF A REINFORCEMENT LEARNING BASED CONTROLLER FOR A FOURTH ORDER FLUID LEVEL SYSTEM
Lucas Matos1; Adolfo Bauchspiess1
1 Universidade de Brasília
Baixar PDF doi:10.20906/CPS/CBA2018-1411
Resumo
This work presents the tuning of an adaptive controller using reinforcement learning and neural networks in order to deal with black-box time-variant nonlinear systems. To evaluate the controller's limitations, a fourth-order fluid level system was chosen because of its wide range of time constants and non-linearities. Implementation was made on a computer running MatLab® connected to an Arduino as interface to the sensor and actuator. The controller was tested with different sample times and different learning rates and, afterwards, was compared to a PI controller. The control-ler was able to perform inside specific learning and sample rate margins and if given time, shows adaptive and opti-mizing features that causes it to perform better than the PI.
Palavras-chave: Reinforcement Learning; Neural Networks; Adaptive Control; Fluid Level Control