# Challenge Data

### EV Charging Stations Usage by Planète Oui

#### Description

##### Dates

Started on Jan. 4, 2021

##### Challenge context

Reducing our overall energy consumption is not a choice anymore ... and towards this, being able to predict when and how we’re using the energy is a crucial knowledge. It lets the energy supplier anticipate the needs and adapt its strategy.

Today electric vehicles are being more and more ordinary, and EV charging stations are consequently being deployed, creating a significant need for electric energy. Being able to predict the usage of these stations and the behavior of the drivers is thus necessary to optimize the energy network. Planète OUI, as an energy supplier, have a great interest in this problematic which falls into the scope of the Consumption Prediction team.

Planète OUI is a French energy supplier, offering to individuals and professionals a 100% renewable energy mix, produced in France through wind turbines, solar panels and hydraulic turbines. Established in 2007 and based in Lyon and Lille, it is today a 120-person company furnishing energy to over 70000 electricity meters. The team operates on all levels of the renewable electricity market: aggregation, production assets management, energy trading, capacity certification, and green electricity supply to the final consumers. Planète OUI is one of the first French green electricity supplier and promotes an green energetic transition for all.

##### Challenge goals

The objective of this challenge is to design a model capable of predicting the usage of some EV charging stations in Paris, more specifically the times when they are available, actively charging a car, plugged, offline or down.

### Terminology

• A terminal is the device on which the car is charging
• Several termninals are grouped together in a station located at a unique physical address
• At each time t, each terminal is in a specific state
• Available: the terminal is unplugged
• Charging: the terminal is being used and is actively charging a car
• Passive: the terminal is plugged but do not deliver energy
• Offline: the terminal is not communicating its state. It may be either available or charging or plugged
• Down: the terminal is in maintenance and not available

Each terminal is identified by its id contructed as follow: S{sid}-T{tid} with {sid} the id of the station and {tid} the terminal number in the station. So all terminals belonging to the same station share the same S{sid} appendix.

### Dataset

This is a timeseries multiclass problem whose objective is to predict the state of each terminal during a period of time. The timseries data is given in Y_train and the candidate can choose to base his model on Y_train only. Some optionnal contextual information are given in X_train, X_test and Static_Info.

• The training set is from November 25th 2019 to November 8th 2020
• The testing set is from November 9th 2020 to November 22th 2020
• All timestamps are in UTC
• Timestamps are expressed under ISO format
• When a state is unknown for a timestamp, the input is NaN
• Be aware of some missing timestamps
• Be aware than the information in X_train are at a step of 1 hour while a step of 15 minutes is required for the predictions

#### Timeseries : Y_train

A dataframe with one column per terminal and one row per timestamps giving the evolution of the status of each terminal over the training period

#### X_train and X_test

X train and X test give additional information:

• The rain level (mm/h)
• The temperature (in Celcius)
• The wind speed (km/h)
• The wind gust speed (km/h)
• The traffic state (Fluid, Busy, Heavy, Jammed)
• The average flow (Number of cars per hour)

#### Static_Info

Static_Info gives some information about each terminal:

• The station it belongs to
• The station coordinates
• The plug models available at this terminal

### Sources

The data has been prepared based on several public datasets:

• Belib' - Prises de recharge pour véhicules électriques - Disponibilité temps réel, Paris City
• Comptage routier - Données trafic issues des capteurs permanents, Paris City
• Live weather forecast, Meteociel

### Metric

The metric used is a Weighted F1 score applied to a multiclass problem.

#### Terminology

• S: states set. S = {Available, Charging, Passive, Offline, Down}
• T: terminals set.
• P{t,s}: Precision for terminal t ∈ T and state s ∈ S
• R{t,s}: Recall for terminal t ∈ T and state s ∈ S
• F1{t,s}: F1 score for terminal t ∈ T and state s ∈ S
• F1{s}: F1 score for state s ∈ S over all terminals
• Prob{s}: Probability of state s ∈ S over the dataset
• F1{final}: Final F1 score

#### Score

For each terminal and state:

$F1_{t,s}&space;=&space;\frac{2*P_{t,s}*R_{t,s}}{P_{t,s}+R_{t,s}}$

For each state:

$F1_{s}&space;=&space;\sum_{T}^{}F1_{t,s}$

Overall:

$F1_{final}&space;=&space;\sum_{S}^{}F1_{s}*Prob_{s}$

### Benchmark

The reference score will be computed by identifying the average behaviour over the week days and the weekend days (without considering additional factors) and applying those behaviours over the period of testing.

#### Files

Files are accessible when logged in and registered to the challenge

Data Scientist