Nutzung von Support Vector Machines für Ortung und Navigation in der Luftfahrt

The purpose of this work is to reproduce the cognitive and learning processes involved in the field of self-localization and the estimation of the current flight state using Machine Learning and Support Vector Machines (SVM) in particular. A core element of many modern localization algorithms are Bayes filters. An iterative process of time and measurement updates allows the continous estimation of an aircraft's current position and state. A necessary prerequisite for this process are highly sophisticated mathematical models that describe the interpretation of sensor readings as well as the propagation of the aircraft's state in time. Human pilots can usually accomplish this task even without explicit know\-ledge of the flight-mechanical and mathematical principles. They can replace it by previously made observations and by experience. In this work, the biologically motivated and model-free learning of functional relationships regarding the interpretation of sensory perceptions and the propagation of an aicraft's state will be mimicked by Support Vector Machines. They usually outperform classical neural networks such as the multilayer-perceptron in the field of classification and regression problems because of their very good capability to generalize. In addition to the design of a filter architecture suited for the incorporation of Support Vector Machines, this work covers the SVM-based correction of raw data from single- and multi-sensor systems, a SVM system for propagating the state vector as well as procedures for the generation of suitable training data. Finally, a localization method is outlined that is able to determine the respective probability density functions for each time and measurement update within the Bayesian filtering process using only basic arithmetic operations.