Development and Evaluation of Adaptive Methods for Fault Detection in Building Operation
In order to ensure fault-free and energy-efficient building operation, it is necessary to acquire and analyze relevant time series data, such as system temperatures, control signals or weather data. For a largely automated analysis of these data with regard to suboptimal operating states / errors, different methods can be used. Promising methods are common regression and classification methods, such as linear regression, support vector machines, decision trees, or naive Bayes classifiers. A prerequisite for the functioning of these and similar methods is the existence of a training data set which covers as far as possible (nominal and faulty) system states. Since this requirement is not fulfilled in practice in most cases, one can pass on to adjust the training data set gradually during operation.
Ort: Raum 04.137, Martensstr. 3, Erlangen