IEEE PVSC 49
Search
SPLTRAK Abstract Submission
Surrogate Modeling for Rapid Prediction of Energy Yield from Vehicle-Integrated Photovoltaics
Timofey Golubev
ThermoAnalytics, Inc., Calumet, MI, United States

Vehicle-integrated photovoltaics (VIPVs) may benefit electric vehicle (EV) performance by extending the driving range, reducing the frequency of charging, and powering secondary electronic systems. Evaluating the electricity production of VIPVs through simulation is necessary to efficiently design these systems and estimate their energy yield. In previous work, we developed a VIPV modeling approach that considers thermal effects by coupling a commercial heat transfer software with temperature-dependent electrical models. While this approach allows for careful consideration of the vehicle’s geometry and the system’s thermal and electrical properties, it takes more than one hour to run a year-long energy production simulation for a single location. For comprehensive studies of energy yield under varying geographical and meteorological boundary conditions, faster simulations are desirable. In this work, we develop an approach for rapid prediction of VIPV energy yield through surrogate modeling. In our approach, the surrogate model is trained using results from physics-based thermal-electrical simulations. The surrogate model is chosen by comparing the performance of elastic net, support vector machine, random forest, gradient boosting, and artificial neural network machine learning algorithms, tuning the hyperparameters for each by conducting grid searches with cross-validation.  An artificial neural network is chosen as the best-performing model. The neural network is found to be able to predict a VIPV system’s energy production with 5 orders of magnitude speed-up and less than 3% error when compared to the physics-based simulation. The surrogate modeling approach developed in this work enables efficiently conducting comprehensive studies of a VIPV system’s energy yield during different seasons and in different regions.