Teil eines Buches,
On Ensemble Classifiers for Nonintrusive Appliance Load Monitoring
Hybrid Artificial Intelligent Systems: 7th International Conference, HAIS 2012, Salamanca, Spain, Proceedings, Part I, 7208, Springer, Heidelberg, (2012)
DOI: 10.1007/978-3-642-28942-2_29
Zusammenfassung
In this work we employ ensemble classifiers for the problem of nonintrusive appliance load monitoring. In practical scenarios the question arises how to efficiently and automatically learn statistical models for appliance recognition, which is an important step for various problems in process recognition, healthcare, and energy consulting. This work is an application study that analyzes multi-class support vector machines (SVMs), and K-nearest neighbors (KNN) in the problem domain of automatically recognizing appliances. By combining two types of classifiers with varying parameterizations to ensembles, we reduce the classification error, and increase the robustness of the classifier. In the experimental part we consider a field study with household appliances, and compare the classifiers w.r.t. various training set and neighborhood sizes. It turns out that the ensembles belong to the best classifiers in all training set scenarios.
