Data-driven forecasting of cross-border power flows across NEMO cable using machine learning and statistical techniques

This study investigates short-term electricity flow forecasting for the NEMO Link interconnector between Belgium and Great Britain. The goal is to support effective grid operation and energy trading. Using a step-by-step approach, the study generated 48-hour-ahead forecasts and evaluated three modelling techniques, namely Autoregressive Moving Average with eXogenous inputs (ARMAX), Long Short-Term Memory (LSTM), and Light Gradient Boosting Machine (LightGBM). The performance of these models was evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and the Coefficient of Determination ( R 2 ). The models were evaluated after feature engineering and careful model selection. The results show that all three models provide accurate predictions over the test period, with R 2 values of at least 0.94. ARMAX achieved the lowest numerical RMSE and MAE over the three-month test period and the lowest numerical errors for the representative 48-hour recursive forecasting window. Statistical comparison using the Diebold–Mariano (DM) and Harvey–Leybourne–Newbold (HLN)-corrected DM tests provided a balanced assessment of model performance across different evaluation periods. In addition, rolling-origin validation demonstrated that forecasting performance varies with the forecasting period while remaining robust across multiple evaluation windows. Bootstrap-based 95% prediction intervals were constructed to quantify forecasting uncertainty. These results highlight the value of combining feature selection, recursive forecasting, rolling-origin validation, uncertainty quantification, and statistical significance testing for reliable short-term interconnector flow forecasting.