TY - GEN
T1 - Modeling of a Motor-driven Propeller Dynamics System by Neural Ordinary Differential Equation
AU - Peng, Chao Chung
AU - Chen, Yi Ho
N1 - Publisher Copyright:
© 2023 IEEE.
PY - 2023
Y1 - 2023
N2 - In recent years, the prosperous development of neural network-related technologies has solved lots of problems in the real world, including text generation, image processing and object recognition. Fewer studies are used for system dynamics modeling, simulation, and identification. To simulate the transient response of a dynamical system, most of the studies applied recurrent neural network (RNN), which uses feedback connections to consider the influence of past state when computing current state. However, it still needs a much larger structure relative to the order of the system dynamics for neural network to have a good fitting performance. To reduce the model complexity and increase the accuracy, this paper presents a data-driven based modeling for a motor-driven propeller system by applying the neural ordinary differential equation (i.e., neural ODE). It is worthy to note that the neural ODE builds a continuous-time model for system identification, which can reduce large amount of memory demands to store the past states generated by the RNN. By comparing with the exact physical model, the discrete-time RNN, and the nonlinear autoregressive with exogenous (NARX) model, numerical simulations show that the proposed neural ODE model gives the best fitting results in the presence of measurement noise.
AB - In recent years, the prosperous development of neural network-related technologies has solved lots of problems in the real world, including text generation, image processing and object recognition. Fewer studies are used for system dynamics modeling, simulation, and identification. To simulate the transient response of a dynamical system, most of the studies applied recurrent neural network (RNN), which uses feedback connections to consider the influence of past state when computing current state. However, it still needs a much larger structure relative to the order of the system dynamics for neural network to have a good fitting performance. To reduce the model complexity and increase the accuracy, this paper presents a data-driven based modeling for a motor-driven propeller system by applying the neural ordinary differential equation (i.e., neural ODE). It is worthy to note that the neural ODE builds a continuous-time model for system identification, which can reduce large amount of memory demands to store the past states generated by the RNN. By comparing with the exact physical model, the discrete-time RNN, and the nonlinear autoregressive with exogenous (NARX) model, numerical simulations show that the proposed neural ODE model gives the best fitting results in the presence of measurement noise.
UR - http://www.scopus.com/inward/record.url?scp=85171472688&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85171472688&partnerID=8YFLogxK
U2 - 10.1109/IS3C57901.2023.00082
DO - 10.1109/IS3C57901.2023.00082
M3 - Conference contribution
AN - SCOPUS:85171472688
T3 - Proceedings - 2023 6th International Symposium on Computer, Consumer and Control, IS3C 2023
SP - 284
EP - 287
BT - Proceedings - 2023 6th International Symposium on Computer, Consumer and Control, IS3C 2023
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 6th International Symposium on Computer, Consumer and Control, IS3C 2023
Y2 - 30 June 2023 through 3 July 2023
ER -