Prognosesysteme für die Verkehrssicherheit mit Methoden des Soft-Computings am Beispiel einer Glätteprognose und einer Fahrzustandsbestimmung

In this work new prognosis systems for road safety are presented. The systems enable the entry and optimization of dynamic behavior in traffic consideration. To get such results an adaptation procedure was developed and tested by optimized weighting values of knowledge base of a fuzzy system. Furthermore in this work a procedure was developed, which is able to extend and optimize the knowledge base of a fuzzy system by genetic algorithms. Thereby it is not necessary to know and describe all connections between the available values. Also the learning algorithm for neural fuzzy networks by using α-cuts was extended with a genetic algorithm. This enables a faster learning of the network and the determination of the optimal results. It was shown that with the help of the described methods errors in measured values can be compensated very well. Furthermore it was shown that an intelligent system is also able to determine not directly measurable values and make with these values sufficiently good prognoses. This was shown in this work with the smoothness prognosis. The implementation of the system shows that the prognosis accuracy is 95%. With the implemented system for the jam prognosis the prognosis accuracy is 75%. This value was increased in the course of the work by inclusion of the drivers behavior to 86%. For use of the measured value this is sufficient. An adaptation, which was implemented in the course of the work, did not bring significant improvement. This was also expected, since the adaptable values are taken off by the user profile mainly. In systems, for different users more difficult optimization tasks have to be implemented, since continuing changing from user profiles prepares large problems. An optimized system created, can represent a perfectly insufficient profile for another user. The beginnings for detecting the driver behavior, selected here, are to be forecastable for the next reactions and are sufficiently exact. Although both procedures are very different, they produce approximately the same results. The resulting membership functions are alike up to small differences. Finally, the determined procedures are not only limited to these applications.