Predictive modelling applied to estimate demand on new EV charging stations in the UK

Sammanfattning: The arrival of the electric vehicle is being faster than estimated. This growth requires a charging infrastructure that can cover the energy needs of electric mobility. To this end, it is necessary that the charging points be installed in the most suitable places to satisfy the coming demand. The optimal and efficient choice of the location of new charging sites will not only help to absorb more demand, but will also improve the economic prospect and facilitate the adoption of electric vehicles. Tesla is the fastest growing electric vehicle company in recent years, with over 500,000 EVs worldwide and some 35,000 in the UK, reporting real-time data on various vehicle parameters. This data is very advantageous for understanding the customer's charging needs from a geographical and demand point of view. This project therefore consists of the development of data tools that allow the precise location and size of new charging points in the UK. A first demand planning tool informs about the number of charging stalls required in a given region covered by an existing Superchargers. A second data tool, built from a Machine Learning model, will predict the number of monthly charging sessions that the new Supercharger will receive. To develop both tools, Tesla data has been examined from different sources describing the geographic characteristics of the charger, driver, and vehicle, and found a regression to an average monthly charge, plus a required number of stalls per region. The thesis is written for Tesla and uses the data sources available to the company. Several machine learning methods have been evaluate to confirm which is the most successful in this analysis. Also, other features are analysed such as, how the frequency of data collection affects prediction, and which characteristics most influence the demand for a new Supercharger. Important characteristics for the demand for a supercharger are related to traffic, number of nearest customers and the type of supercharger. The demand planning shows that the UK will need more than 200 new stalls to cover the demand of December 2021. In addition, the charging sessions prediction tool has an average error of 18% which will allow to optimize the distribution of new superchargers. Finally, this tool is used by the charging department which together with the knowledge of the project developer allows to make an optimal choice of the location of a new supercharger.