Virtual Strategy Engineer: Using Artificial Neural Networks for Making Race Strategy Decisions in Circuit Motorsport

Heilmeier, Alexander; Thomaser, André; Graf, Michael; Betz, Johannes

doi:10.3390/app10217805

Cited by 15 publications

(10 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…An illustrative and graphic representation of the pit stop strategy problem is presented in Chain Bear (2017). In a recent paper, Heilmeier et al (2020) use neural networks (NN) calibrated on real data to predict if a pit stop should be made during the current lap, and the new tire compound to be used. By computing the optimal race strategy under a no-competition setting with a mixed-integer quadratic program (MIQP), the authors integrate these NN for all drivers in a simulation to determine the race outcome.…”

Section: Related Workmentioning

confidence: 99%

On the optimization of pit stop strategies via dynamic programming

Heine

Thraves

2022

Cent Eur J Oper Res

View full text Add to dashboard Cite

Pit stops are a key element of racing strategy in several motor sports. Typically, these stops involve decisions such as in which laps to stop, and which type of tire, of three possible compounds, to set at each of these stops. There are several factors that increase the complexity of the task: the impact of lap times depending on the tire compound, the wear of the tires, unexpected events on the track such as safety cars and the weather, among others. This work presents a Dynamic Programming formulation that addresses the pit-stop strategy problem in order to optimize the laps in which to stop, and the tire changes that minimize the total race time. We show the relative performance of the optimal strategies for starting with tires of different compounds with different yellow-flag scenarios. Then, we extend the Dynamic Program (DP) to a Stochastic Dynamic Programming (SDP) formulation that incorporates random events such as yellow flags or rainy weather. We are able to visualize and compare these optimal pit-stop strategies obtained with these models in different scenarios. We show that the SDP solution, compared to the DP solution, tends to delay pit stops in order to benefit from a possible yellow flag. Finally, we show that the SDP outperforms the DP, especially in races in which yellow flags are likely to be waved more frequently.

show abstract

Section: Related Workmentioning

confidence: 99%

On the optimization of pit stop strategies via dynamic programming

Heine

Thraves

2022

Cent Eur J Oper Res

View full text Add to dashboard Cite

show abstract

“…The importance of the FCY phase on race results is evident from the championship-deciding race 2021 Abu Dhabi Grand Prix, where the choice of pitstop before and during safety car changed the race results. This FCY phase is discussed in detail by Heilmeier and Thomaser et al [7].…”

Section: A Formula One Racing Terminologies and Rulesmentioning

confidence: 93%

Deep-Racing: An Embedded Deep Neural Network (EDNN) Model to Predict the Winning Strategy in Formula One Racing

2023

IJML

View full text Add to dashboard Cite

This paper presents an embedded deep neural network model to predict the driver rank and the optimum pitstop strategy. In formula one racing, the race strategy is critical to determine optimal pitstops and finish the race in the best possible position. Considering a system with only one racing car, the pitstop can be decided just by looking at the degradation of the tires. But in reality, the formula one environment is more complex, and multiple probabilistic factors (like safety car phases, opponent strategy, and overtaking) influence the pitstop decision. Deep-Racing is a prediction and decision algorithm for formula one racing cars that uses neural networks with embedding layers. The algorithm is developed after carefully reviewing formula one racing and appropriate statistical modeling techniques, which can be trained for pre-race and real-time predictions during the race using the data from previous laps. Deep-Racing has the potential to help team principals and race engineers to decide the optimized strategy for making pitstops. It is trained on the data from seasons 2015-2022. This project is the first to utilize an embedded layer in motorsport racing predictions, and the results show an improvement in predictive accuracy compared with the previously available literature. This paper significantly expands the previous research in this field and proposes trends in the data available from the latest seasons. Index Terms-Neural networks applications, formula one racing, embedded deep neural networks, predictive analysis I. INTRODUCTIONFormula one racing has the world's most extensive viewing of all the circuit motorsport, with a combined viewership of 445 million in 2021 [1, 2]. Formula one cars are a marvel of engineering and have been considered among the fastestdesigned vehicles. Formula one, or simply F1, is sanctioned by Fédération Internationale de l'Automobile (FIA), which sets the rules and regulations for the races, including the car design, frequency of pitstops, and choice of tires. Multiple teams, including Ferrari, Mercedes, and McLaren, have participated in the world championship every year since its inaugural season in 1950, involving a series of Grand Prix [3]. Grand Prix is a race performed on closed road circuits of varying lengths and complexities across various parts of the world, including Monaco, Abu Dhabi, Singapore, Monza, and Silverstone. F1 racing dominates the betting sites where sportsbooks and fans are interested to know the prediction of the winner before the race day. The goal of a team in F1 racing is simple -be the first one to cross the chequered flag. The race result depends not only on the driver or the car's pace but also, on multiple real-time stochastic elements such as the timing and duration of pitstops, tire degradation, and weather conditions. Although the driver's skills and the car's Manuscript

show abstract

“…The aforementioned works also consider the behavior of the competitors. Moving towards decision-making, a data-driven ANN was proposed to decide on the tire strategy [36]. Gametheoretical approaches for planning overtake maneuvers were explored for autonomous race cars [37] and sailing yachts [38].…”

Section: A Literature Reviewmentioning

confidence: 99%

“…At this point, the energy management strategy in terms of fuel deployment is already as aggressive as it can be and the quantity of fuel that can be burned by keeping the engine running during braking maneuvers is limited by the brake balance constraints (see Section IV-D). However, the fuel allocation has to be completely used up by the end of the lap, due to the equality constraint (36). The optimal solution thus compromises lap time in order to fulfill the fuel constraint, e.g., by delaying the MGU-K power deployment to achieve a lower velocity on the straights, meaning that more time passes during which fuel can be burnt.…”

Section: Lap Time Mapsmentioning

confidence: 99%

Minimum-Race-Time Energy Allocation Strategies for the Hybrid-Electric Formula 1 Power Unit

Duhr¹,

Buccheri²,

Balerna³

et al. 2023

IEEE Trans. Veh. Technol.

View full text Add to dashboard Cite

The hybrid-electric powertrain currently used in Formula 1 race cars draws its energy from the car's fuel tank and battery. The usable battery size is limited, and refueling during a race is forbidden by the regulations of the Formula 1 race series. From a strategic point of view, lap-by-lap targets for the fuel and battery consumption must be chosen and imposed on the energy management controller of the car. This task is non-trivial due to the influence of the on-board fuel mass on the achievable lap time, as well as the cross-couplings between the electric and the combustion part of the powertrain. A systematic approach is thus required to compute the energy allocation strategy that minimizes the total race time. In this paper, we devise an optimization framework in the form of a non-linear program, yielding the optimal battery and fuel consumption targets for each lap of the race. The approach is based on maps that capture the achievable lap time as a function of car mass and allocated battery and fuel energy. These maps are generated beforehand with a model-based single-lap optimization framework and fitted using artificial neural network techniques. To showcase the approach, we present three case studies: First, we compare the optimal strategy to a heuristic method. The improvement of 2 s over the entire race is substantial, given that the difference only lies in the energy allocation, but not in the overall consumption. It underlines the importance of optimizing the energy allocation. Second, we leverage the framework to compute the optimal fuel load at the beginning of the race. Finally, we apply the developed non-linear program in a shrinking-horizon fashion. Our simulation results show that the resulting model predictive controller correctly reacts to disturbances that frequently occur during a race.

show abstract

Virtual Strategy Engineer: Using Artificial Neural Networks for Making Race Strategy Decisions in Circuit Motorsport

Cited by 15 publications

References 29 publications

On the optimization of pit stop strategies via dynamic programming

On the optimization of pit stop strategies via dynamic programming

Deep-Racing: An Embedded Deep Neural Network (EDNN) Model to Predict the Winning Strategy in Formula One Racing

Minimum-Race-Time Energy Allocation Strategies for the Hybrid-Electric Formula 1 Power Unit

Contact Info

Product

Resources

About