Environmental Parameter Prediction for Connected Cars using Machine Learning

Many functionalities in modern production cars need information about environmental parameters like temperature, solar radiation or road friction coefficients. These information are used within different modules, e.g. in the driving and operating strategies or for range prediction. This implies that current and future values of these environmental parameters have to be known to obtain optimal and accurate results for the single modules. In general, the environmental parameters are measured by the vehicles themselves. By collecting the information of multiple connected vehicles in a database, it is possible to obtain spatial and temporal distributed measurements of the environmental parameters. Based on these measurements, a prediction procedure for the environmental parameters is developed to obtain future values of them at specific route sections and instants of time, which can be used in predictive operating strategies and for range prediction. For the forecasts of the environmental parameters, different machine learning algorithms are deployed and an algorithm for input selection is applied to reduce the model complexity. The proposed procedure is validated using recorded temperature measurements.