• Mikkel Giedsing Nielsen
  • Jacob Elefsen
4. semester, Software, Kandidat (Kandidatuddannelse)
This report considers the problem of predicting a trustworthy energy consumption for a future trip. To this end, we have access to a data warehouse that contains historical observations about trips that are map-matched to a road network. We construct a general method of combining a road network with elevation data. The elevation information and other factors which impact the energy consumption are analyzed. From the analysis, a feature set that describes energy consumption is constructed. This feature set does not include information which is not available for future trips, such as the exact speed and time. We construct three models, Linear Regression (LR), Neural Network (NN), and a Neural Network that is combined with historical observations (NN-observations). We evaluate the quality of the feature set using the LR and NN. In order to determine the performance of these three models, we compare them to three baselines. The LR and NN models outperform the baselines by a slight margin. However, we find that the including the historical observations in the NN-observation model increases the performance significantly.
SprogEngelsk
Udgivelsesdato2017
Antal sider75
ID: 259243009