A novel strategy model approach for the Bridgestone World Solar Challenge

Jeremy Budd, Ricardo Enguiça, Yitaek Kim, Julian Koellermeier, S. van Mourik, Arijit Sarkar, Filipa Soares, Nele Thomsen2024 · DOI

Abstract

This project considers the derivation of strategies for a solar car race through Australia, posed by TopDutch Solar Racing. The task is to optimize the arrival time of an electric car powered by solar energy by changing the velocity along the track taking into account the characteristics of the car, external influences such as wind and sun, as well as constraints like the maximal and minimal battery charge. To address this challenge, the problem is modeled and four different optimization strategies are devised, focusing on different time horizons and methods. We explain a data-based long-term strategy, an optimisation-based long-term strategy, a dynamic programming long-term strategy, and an optimization-based short-term strategy. While successful short-term strategies require more work, the long-term strategies give realistic results and are powerful tools to increase the winning chance for the team. We outline several possibilities to improve the models based on additional data, modeling and methods.

Introduction

The Bridgestone World Solar Challenge (BWSC) represents the pinnacle of solar-powered endurance racing, spanning 3,020 km from Darwin to Adelaide. For teams like TopDutch Solar Racing (TDSR), the competition is less about raw speed and more about sophisticated energy management. With a battery capacity of only 5kWh—approximately 1/15th of a standard Tesla Model 3—and strict racing hours (08:00 to 17:00), the strategy must account for road inclination, wind resistance, and variable solar irradiance. The core objective is to determine an optimal velocity profile $v(t)$ that minimizes total race time without depleting the battery below critical levels before reaching a charging window or checkpoint.

Methodology

The researchers developed a hierarchical decision framework to address the multi-scale nature of the race. This framework separates global planning from local adjustments to maintain computational efficiency and robustness.

Hierarchical optimization framework for solar car racing strategy.

1. Data-Driven Modeling

A fundamental component of the strategy is the vehicle's power consumption model. Using performance data from the TDSR 2019 vehicle, a regression model was established to relate motor power ($P$) to vehicle velocity ($v$). The resulting trendline, $P(v) = 0.17v^2 - 6.6v + 252.95$, serves as the baseline for energy expenditure calculations.

Regression model of motor power consumption versus vehicle velocity.

2. Optimization Strategies

Four distinct strategies were evaluated:

  • Data-based Long-term Strategy: A heuristic approach using historical solar data to find a constant optimal velocity.
  • Optimization-based Long-term Strategy: Utilizing Particle Swarm Optimization (PSO) to generate non-constant velocity profiles based on objectives like maximizing finishing probability or minimizing expected time.
  • Optimization-based Short-term Strategy: A Model Predictive Control (MPC) approach designed for a 30-60 minute horizon to react to immediate environmental disturbances.
  • Dynamic Programming (Stochastic Control): A method that incorporates Gaussian noise for solar input and model errors to maximize expected daily distance through dynamic feedback.

Results

The long-term optimization model identified a baseline optimal constant velocity of 81 km/h, yielding a finish time of roughly 4.67 days with a 6.34% battery safety margin. Sensitivity analysis revealed a "probability cliff": increasing speed by a mere 1-2 km/h can plummet the probability of finishing from near-certainty to zero, emphasizing the need for high-precision modeling.

Furthermore, the stochastic dynamic programming approach demonstrated the superiority of feedback-based control. By adjusting velocity based on the real-time battery state, the expected daily distance increased from 606 km to 621 km, while simultaneously reducing the variance in the final battery state. This increased stability allows the team to operate closer to the car's physical limits with reduced risk of failure.

Conclusion and Future Work

The proposed multi-tiered framework provides a robust foundation for solar race strategy. While the long-term strategies produced realistic results consistent with top-tier BWSC performance (79–88 km/h), the short-term MPC strategy requires further refinement to improve solver reliability under velocity regularization. Future developments will focus on integrating real-time telemetry to adaptively tune system dynamics and potentially co-optimizing vehicle design parameters with operational strategy.

Summary updated

Abstract

This project considers the derivation of strategies for a solar car race through Australia, posed by TopDutch Solar Racing. The task is to optimize the arrival time of an electric car powered by solar energy by changing the velocity along the track taking into account the characteristics of the car, external influences such as wind and sun, as well as constraints like the maximal and minimal battery charge. To address this challenge, the problem is modeled and four different optimization strategies are devised, focusing on different time horizons and methods. We explain a data-based long-term strategy, an optimisation-based long-term strategy, a dynamic programming long-term strategy, and an optimization-based short-term strategy. While successful short-term strategies require more work, the long-term strategies give realistic results and are powerful tools to increase the winning chance for the team. We outline several possibilities to improve the models based on additional data, modeling and methods.
BibTeX
@misc{budd2024novel,
  author = {Jeremy Budd and Ricardo Enguiça and Yitaek Kim and Julian Koellermeier and S. van Mourik and Arijit Sarkar and Filipa Soares and Nele Thomsen},
  title = {{A novel strategy model approach for the Bridgestone World Solar Challenge}},
  year = {2024},
  doi = {10.33774/miir-2024-9b1s4},
}
Figure