An AI-Based Speed Forecasting Algorithm for Efficient Energy Management in Hybrid Electric Vehicles

Nowadays, powertrain hybridization represents one of the most viable solutions to reduce the carbon footprint of the transport sector. Nevertheless, its potential can be fully exploited only through the design of suitable Energy Management Strategies capable of optimally coordinating the multiple power sources installed onboard. In this context, accurate speed forecasting, enabled by increasing levels of vehicle connectivity, can support the development of predictive control strategies that further enhance the benefits of hybridization through a more efficient optimization of energy flows. Therefore, this thesis work investigates the use of predictive artificial intelligence for energy management enhancement through vehicle speed forecasting. Specifically, the study employs a Long Short-Term Memory (LSTM) neural network trained on synthetic traffic data generated with the open-source software SUMO (Simulation of Urban Mobility), simulating a typical working day in the city of Turin. The dataset was derived from SUMO’s mesoscopic mode and processed through a Python pipeline for feature extraction and MATLAB scripts for filtering, normalization, and LSTM sequence preparation and for the training of the networks themselves. Several network configurations were trained and evaluated using RMSE and MAE metrics to identify the optimal hyperparameters. The best-performing model, trained on data from 2000 vehicles, achieved an average RMSE of approximately 0.283 and an MAE of 0.240 on 80-second sequences, demonstrating the ability to reproduce general speed trends with satisfactory accuracy. Additional tests on 140-second sequences confirmed the model’s capacity to capture temporal dependencies, though challenges remain for sequences with weak time correlations. Future developments may focus on extending the dataset to microscopic simulations to better reflect real urban driving behavior and further enhance prediction accuracy for energy optimization purposes.