Modeling of Total Occupancy Curves with Integrated Single Parking Times as Input for Microscopic Traffic Simulation

This paper presents a novel approach for the preparation of real parking data for microscopic traffic simulation or similar applications. Against the backdrop of an incomplete database given from single parking spot sensors and manual countings on a few parking spaces in Munich, the data are cleaned and expanded by various methods, e.g. a neural network to forecast the total occupancy at each hour as well as an optimization model to generate the corresponding parking times of each car. This results in a list of realistic parking events for a certain location and day. The method is transferable to other use cases regarding off-street parking spaces, particularly with an incomplete and discontinuous detection of parking occupancy and times.