Active Control of a Multivariable System Via Polynomial Neural Networks: Computer Simulation Evaluations and Laboratory Experimental Results
Abstract
The objective of the work described herein was to develop, implement, and demonstrate inductively synthesized active control algorithms that minimize a performance metric that is a function of a signal measured by a sensor external to a multivariable control system. The controller is required to control a (potentially nonlinear) plant when subjected to a broadband (impulsive) disturbance signal. Polynomial neural networks (PNNs) are used to implement the control algorithms. To provide controller/secondary feedback compensation filter parameters for the laboratory experiments, and to provide a benchmark with which to compare the empirical results, computer simulation evaluations were also conducted. The report documents controller designs that achieve up to 24.4 dB error improvement relative to the uncontrolled case. Key differences between the present work and that ongoing elsewhere in active control are: (1) The use of broadband impulsive disturbances, which induce a more complex stochastic control scenario than that of narrowband excitation; (2) The potential controller structures are adaptive, nonlinear, infinite impulse response PNNs, not traditional linear, finite impulse response filters; the former offer promising new active control approaches for non-Gaussian, as well as Gaussian, signals, and may be used for multiple-input, multiple-output control of unknown plants. Active control, Polynomial neural networks, Feedforward control, Broadband control, Predictive-feedback control, Nonlinear control, Multivariable control, Computer simulations.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jun 01, 1993
- Accession Number
- ADA270033
Entities
People
- B. E. Parker Jr.
- David G. Ward
- Natalie A. Nigro