Sport analytics is a promising area of research and application of data science solutions, that aims to gain insights, make informed decisions and improve performance in the field of sports. In particular, Formula 1 is one of the sport having the highest potential in terms of applications of data analytics tool for strategic predictions and decision-making support. One of the key factors influencing race performance is the tire, standardized in size and strategic in compound selection. This constitutes a significant challenge for Pirelli Motorsport team, consisting of experts responsible for the entire Formula 1 tire domain. They provided supervision for this project, endorsing it, providing a substantial database constituted by measured and estimated data. The primary objective of this thesis is to identify similarities between different tracks, in order to support strategies formulation in similar contexts, then predict lap time degradation and the main features influencing it. To achieve this goal, RACER-KIT was developed: a model based on the k-means algorithm utilizing a database encompassing both intrinsic features and indicators estimated by experts. The results, comparing two consecutive years, namely 2021 and 2022, are promising and demonstrate satisfactory accuracy. RACER-KIT also supports SWIFT, a predictive algorithm aiming to forecast lap time degradation, one of the most important tire behavior effects. SWIFT is an adaptive ensemble model, able to combine the predictions of three base learners (i.e., naïve approach, linear models, and XGBoost) on the basis of the current scenario. Moreover, SWIFT supports the on-demand extension of the training set (leveraging RACER-KIT) to include data coming from similar races. The efficiency of SWIFT yields results in the order of microseconds, making it applicable even in real-world contexts.
Sport analytics è un’area di ricerca relativa al data science particolarmente promettente che mira ad ottenere approfondimenti, prendere decisioni coscienti e migliorare le prestazioni nel campo sportivo. In particolare la Formula 1 (F1) è uno degli sport che meglio si presta in termini di applicazione di strumenti di data analytics per previsioni strategiche e supporto decisionale. Uno degli elementi che guida l’andamento di gara è proprio lo pneumatico, standard nella struttura e strategico nella scelta della mescola. Ciò costituisce una sfida significativa per il team Pirelli Motorsport, formato da esperti che gestiscono l’intero mondo degli pneumatici di F1 ed hanno supervisionato interamente questo progetto, avvalorandolo. Il primo obbiettivo di questa tesi è trovare la similarità tra piste differenti, così da poter raffinare le strategie in contesti simili, e predire il successivo tempo sul giro, ricavando le caratteristiche che maggiormente influenzano questa analisi. Con questo obbiettivo RACER-KIT è stato implementato: un modello basato su l’algoritmo k-means che sfrutta un database che comprende sia caratteristiche intrinseche dei circuiti che indicatori stimati dagli esperti. Il risultato, confrontando due anni consecutivi, quali 2021 e 2022, è stimolante e di soddisfacente accuratezza. RACER-KIT può essere integrato come supporto a SWIFT, un algoritmo predittivo che mira a prevedere il degrado in termini di tempo sul giro, consapevoli che una delle sue implicazioni sia proprio il comportamento dello pneumatico. SWIFT è un modello ensemble adattivo, in grado di combinare le previsioni di tre base learner (approccio naïve, modello lineare e XGBoost) sulla base dello scenario attuale. Inoltre, SWIFT supporta l’estensione on-demand dell’insieme di addestramento, sfruttando RACER-KIT per includere dati provenienti da gare simili. Il risultato di SWIFT è valutabile nell’ordine di microsecondi e per la sua efficienza può essere applicato anche in contesti reali.
RACER-KIT and SWIFT models for Pirelli: an advanced algorithmic approach for track classification and lap time degradation prediction
Guerini, Cecilia
2022/2023
Abstract
Sport analytics is a promising area of research and application of data science solutions, that aims to gain insights, make informed decisions and improve performance in the field of sports. In particular, Formula 1 is one of the sport having the highest potential in terms of applications of data analytics tool for strategic predictions and decision-making support. One of the key factors influencing race performance is the tire, standardized in size and strategic in compound selection. This constitutes a significant challenge for Pirelli Motorsport team, consisting of experts responsible for the entire Formula 1 tire domain. They provided supervision for this project, endorsing it, providing a substantial database constituted by measured and estimated data. The primary objective of this thesis is to identify similarities between different tracks, in order to support strategies formulation in similar contexts, then predict lap time degradation and the main features influencing it. To achieve this goal, RACER-KIT was developed: a model based on the k-means algorithm utilizing a database encompassing both intrinsic features and indicators estimated by experts. The results, comparing two consecutive years, namely 2021 and 2022, are promising and demonstrate satisfactory accuracy. RACER-KIT also supports SWIFT, a predictive algorithm aiming to forecast lap time degradation, one of the most important tire behavior effects. SWIFT is an adaptive ensemble model, able to combine the predictions of three base learners (i.e., naïve approach, linear models, and XGBoost) on the basis of the current scenario. Moreover, SWIFT supports the on-demand extension of the training set (leveraging RACER-KIT) to include data coming from similar races. The efficiency of SWIFT yields results in the order of microseconds, making it applicable even in real-world contexts.File | Dimensione | Formato | |
---|---|---|---|
2024_04_Guerini_ExecutiveSummary.pdf
non accessibile
Descrizione: Executive Summary
Dimensione
654.15 kB
Formato
Adobe PDF
|
654.15 kB | Adobe PDF | Visualizza/Apri |
2024_04_Guerini_Tesi.pdf
non accessibile
Descrizione: Tesi: RACER-KIT and SWIFT models for Pirelli: an advanced algorithmic approach for track classification and lap time degradation prediction
Dimensione
1.3 MB
Formato
Adobe PDF
|
1.3 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/217927