Online Reduced-Order Kernel Regression for Data Processing in Sensor Network

摘要

Online data processing in sensor networks problem is of particular relevance in connection with the widespread use of such systems for restoring physical fields from local measurements, for example, in environmental monitoring systems. In order to reduce model dimensionality, clustering of sensors net is carried out previously. In this paper, we consider the problem of restore fields of complex structure using data from sensor network. Nonlinear distributed regression kernel-based method is applied using sequential regularization based on optimal estimates concordance. A generalization of this approach to non-stationary case is also considered, for which recurrent regularized kernel-based learning algorithms are proposed, using both moving average and sliding window approaches.