TY - GEN
T1 - Data-driven tracking controls of multi-input augmented systems based on ADP algorithm
AU - Lv, Yongfeng
AU - Ren, Xuemei
AU - Hu, Shuangyi
AU - Li, Linwei
AU - Na, Jing
N1 - Publisher Copyright:
© 2019 IEEE.
PY - 2019/5
Y1 - 2019/5
N2 - The data-driven optimal tracking controls (OTC) for the unknown multi-input system are proposed in this paper, and a novel tuning law is used to update NN weights in the learning scheme. First, the formula of the OTC for the multi-input NZS game is presented. A three-layer neural network (NN) data-driven model is introduced to approximate the unknown system, and the input dynamics are obtained. Then, to solve the OTC as a regulation optimal problem, an augmentation multi-input system is constructed with the tracking error and command trajectory. Moreover, we use a reinforcement learning based data-driven NN method to online learn the optimal value functions for each input, which is directly used to calculate the optimal tracking control associated with each performance index function. The convergence of the NN weights is proved. Finally, a simulation is presented to verify the feasibility of our algorithm in this paper.
AB - The data-driven optimal tracking controls (OTC) for the unknown multi-input system are proposed in this paper, and a novel tuning law is used to update NN weights in the learning scheme. First, the formula of the OTC for the multi-input NZS game is presented. A three-layer neural network (NN) data-driven model is introduced to approximate the unknown system, and the input dynamics are obtained. Then, to solve the OTC as a regulation optimal problem, an augmentation multi-input system is constructed with the tracking error and command trajectory. Moreover, we use a reinforcement learning based data-driven NN method to online learn the optimal value functions for each input, which is directly used to calculate the optimal tracking control associated with each performance index function. The convergence of the NN weights is proved. Finally, a simulation is presented to verify the feasibility of our algorithm in this paper.
KW - Approximate dynamic programming
KW - Data driven control
KW - Multi-input system
KW - Nash equilibrium
KW - Neural network
UR - http://www.scopus.com/inward/record.url?scp=85076443803&partnerID=8YFLogxK
U2 - 10.1109/DDCLS.2019.8909070
DO - 10.1109/DDCLS.2019.8909070
M3 - Conference contribution
AN - SCOPUS:85076443803
T3 - Proceedings of 2019 IEEE 8th Data Driven Control and Learning Systems Conference, DDCLS 2019
SP - 534
EP - 538
BT - Proceedings of 2019 IEEE 8th Data Driven Control and Learning Systems Conference, DDCLS 2019
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 8th IEEE Data Driven Control and Learning Systems Conference, DDCLS 2019
Y2 - 24 May 2019 through 27 May 2019
ER -