Event-Based Control and Estimation with Stochastic Disturbances
Sammanfattning: This thesis deals with event-based control and estimation strategies, motivated by certain bottlenecks in the control loop. Two kinds of implementation constraints are considered: closing one or several control loops over a data network, and sensors that report measurements only as intervals (e.g. with quantization). The proposed strategies depend critically on _events_, when a data packet is sent or when a change in the measurement signal is received. The value of events is that they communicate new information about stochastic process disturbances. A data network in the control loop imposes constraints on the event timing, modelled as a minimum time between packets. A thresholdbased control strategy is suggested and shown to be optimal for firstorder systems with impulse control. Different ways to find the optimal threshold are investigated for single and multiple control loops sharing one network. The major gain compared to linear time invariant (LTI) control is with a single loop a greatly reduced communication rate, which with multiple loops can be traded for a similarly reduced regulation error. With the bottleneck that sensors report only intervals, both the theoretical and practical control problems become more complex. We focus on the estimation problem, where the optimal solution is known but untractable. Two simplifications are explored to find a realistic state estimator: reformulation to a mixed stochastic/worst case scenario and joint maximum a posteriori estimation. The latter approach is simplified and evaluated experimentally on a moving cart with quantized position measurements controlled by a low-end microcontroller. The examples considered demonstrate that event-based control considerably outperforms LTI control, when the bottleneck addressed is a genuine performance constraint on the latter.
KLICKA HÄR FÖR ATT SE AVHANDLINGEN I FULLTEXT. (PDF-format)