RINA - Démonstrateur d’utilisation de l’intelligence artificielle pour la gestion opérationnelle des risques naturels gravitaires géologiques | FEREC 2020

rock fall. Only the 3 stations along the road are taken into account as they are the closest to the road. The temperature was only available from 2009 onwards. The correlation between both stations was 0.97 for the max and min temperature so the values of Saint-Denis were used. Rock falls were on a table with a list of events with the description of each event. For the majority of the models the objective was to predict if there was an event or not, so a table of days was formed with the data and a 1 was indicated for days with a rockfall event or a 0 if there wasn’t. 3. Preprocessing, data description and previous work The analysis will treat the problem temporally, but a spatial understanding could be useful for a better understanding of the behaviour of the cliff. The cliff did not change significantly during the studied period. The only changes that will be considered were the structures built in 2008 to protect the road as they had a significant influence on the probability of blocks hitting the road. This leads us to study the periods before and after the construction as 2 different data sets. The data can be studied in 3 different formats: temporally, spatially or both. Rock falls occur on a certain time and a certain location and statistical models can help to predict both of them. This study will only study the rock falls in function of the time. Being able to predict the date of the event permits its temporary closure in days the risk is higher in order to avoid accidents. A model that combines spatial and temporal elements would not be ideal as the data set contains few events and adding dimensions would increase this problem. Even though the predictive models will be used for the predictive models a spatial study was made to better understand the cliff. Different descriptive values were calculated to try to understand which characteristics of this bedrock increased the risk of blocs detaching and reaching the road. The time set for each entry is on the order of the day. This permits an easy comparison with the operational code as the closure of the road is dictated by the behavior of the rain accumulated in a period of 24h. Grouping the data by month would facilitate finding tendencies but would make it harder to apply the model for the operation of the road. In this study, we used existing data (1- 4) and had to calculate other (5 -7) : 1. Information about the rock falls that occurred between 1999 and 2018. For each event the table gave the date and time of the event, the point it fell along the road (PR), the lanes affected, the mass that fell and the consequence of the event. 2. Values of the daily rain from 2000-2020 collected on three different stations (Figure 3) 3. Location of the protective elements along the road before and after the constructions of 2008. 4. Geological characteristics of the bedrock along the road 12