Clustering in Supervised Multi-model Adaptive Control Applied to Neuromuscular Blockade

Abstract

This work approaches the problem of forming clusters of linearized models that cover the possible dynamic behavior of neuromuscular blockade of patients subject to anaesthesia. The motivation stands from the design of supervised multi-model adaptive controllers where a group of “similar” models is associated with a single controller. Due to this connection to a control problem, the similarity among models is measured using the ν – gap metric since this ensures that, if two models are close in this norm, then a controller that stabilizes one of the models will also stabilize the other. Two algorithms are proposed for model clustering. The first one starts with an initial classification that relies on insight from the particular problem of neuromuscular blockade control. The other may be used in other applications as well and relies on an initialization based on agglomerative clustering techniques. In both cases, the initial classification is then improved by the k – means algorithm.