Hybrid SARIMA+BO-LSTM Framework for Forecasting EV Adoption: A Road to Net-Zero in Ireland

Abstract

Reducing greenhouse gas (GHG) emissions from the transport sector is central to achieving Ireland’s national climate goals. To support the Climate Action Plan target of registering 945,000 electric vehicles (EVs) by 2030, this study develops a hybrid time series forecasting framework that combines a Seasonal Autoregressive Integrated Moving Average (SARIMA) model with a Bayesian Optimized Long Short-Term Memory (BO-LSTM) network. SARIMA captures linear and seasonal patterns in monthly EV registration data, while BO-LSTM models the non-linear residual structure. Monthly data from January 2010 to October 2024, sourced from the Society of the Irish Motor Industry (SIMI), is used for model training and evaluation. The SARIMA–BO-LSTM model achieves a Mean Absolute Error (MAE) of 742.99, Root Mean Squared Error (RMSE) of 1200, and R2 of 0.93, outperforming several baseline statistical and machine learning models. Scenario-based forecasts are conducted under three conditions: Business-as-Usual, Accelerated Adoption, and Saturation Bound. Projections show that under the Business-as-Usual scenario, Ireland is likely to fall short of the 2030 EV target. In contrast, the Accelerated Adoption scenario meets the target through sustained exponential growth backed by strong policy and infrastructure investment. The Saturation Bound scenario also reaches the target, though with slower growth after an initial surge due to behavioral and economic constraints. The proposed forecasting framework provides a basis for planning infrastructure, incentives, and regulatory measures consistent with Ireland’s decarbonization objectives.