Iterative Reduction of Market Imbalances with DL for Run-of-the-River Hydropower Plants: a Case Study

This work presents a deep learning (DL)-based forecasting system to reduce imbalances from non-dispatchable run-of-the-river hydropower plants (RoR HPPs) on electricity markets. Applied to the “Ala,” “Bussolengo,” and “Chievo” RoR HPPs cascade (119 MW total capacity in Northern Italy), it implements a two-step iterative approach: long-term Intra-Day (ID) market corrections and short-term XBID fine-tuning. For both forecast horizons, three forecasting approaches were compared - machine learning, state-of-the-art DL with a pretrained foundational model (Chronos-2) in zero-shot and finetuned configuration - across hourly (2023-2024) and quarterhourly (2025) market granularities. Fine-tuned Chronos-2 delivered superior accuracy and stability. Economic analysis showed the iterative strategy outperforming single-horizon forecasts, while mitigating XBID sessions price and liquidity fluctuations by shifting the larger share of corrections toward the more s table I markets. Moreover, higher market granularity amplifies imbalance reduction benefits, underscoring the value of advanced forecasting for non-dispatchable RES integration.