Predicting data from driving profiles

In 2007 vehicles emission accounted for 39.2% of total emissions, contributing negatively to climate change, producing air pollution and noise. Most of this energy comes from oil, so transport is responsible for a large part of greenhouse emissions. Other sectors such as energy production and industry have reduced these gases since 1990. Unfortunately, those related to transport have increased. For this reason, research and development of electric vehicles (EVs) is being carried out. EVs are less polluting as they use electricity and not oil; they can be powered by an external source which supplies electricity or can be autonomous by having batteries, solar cells or an electric generator installed. This thesis is focused on the prediction of driving profiles in hybrid vehicles. Hybrid vehicles (HEVs) are powered by two different engines; a combustion engine and an electric one. Like this, the vehicle can use or alternate both energy sources to move in a more economical and sustainable way without losing performance. Driving profile data is collected and then analysed. Initially, some statistical models like ARMA and exponential smoothing were developed to see if it was enough to predict future velocity. Afterwards, neuronal network was implemented. It is more complex and needs more memory than statistical mothers, but also more efficient. Firstly, very simple model was compiled and in next simulations parameters were changed to see how they affected the response, until the best model was created. To do all this predictive analysis, Python was used. Particularly, to simulate statistical models the following libraries were employed; Statsmodels and Scikit-learn. Regarding to Neural Networks, as I am a beginner in Machine learning, I used Keras and TensorFlow as backend, which are very simple but at same time very complete