Signal-Conditioned Bidding and Curtailment for Wind Power: Under the Nordic Imbalance Settlement Model

Sammanfattning

Wind generation’s growing share of Nordic electricity supply has made imbalance settlement an increasingly costly exposure for producers. Under the Nordic single-price mechanism, all residual imbalance is settled at a single imbalance price, regardless of direction, determined by the mFRR activation market. Before that price is known, the producer fixes a day-ahead position, adjusts on the continuous intraday market, and may curtail output via a setpoint. Whatever gap remains between the committed position and actual output is settled at the imbalance price, with a symmetric eSett fee on every megawatt-hour of imbalance. The incremental revenue from each commitment is determined by the spread between the day-ahead or intraday price and the imbalance price. The central finding is that exploiting the stage-specific spread signal simultaneously raises mean daily revenue and reduces tail loss, with no Pareto trade-off at any stage where a directional spread signal is present. Across three Swedish parks, the proposed strategy raises mean daily revenue by approximately 9–18% and compresses CVaR tail loss by roughly 60% at the most exposed park, with the worst-decile expected revenue at another flipping from a small loss to net positive. All headline magnitudes in this thesis (per-park objective lifts, revenue improvements, and tail-loss reductions) are sourced from the combined summary table. At the day-ahead stage, the spread carries no interval-level directional information, and no rule improves over the forecast baseline. When the intraday price is high relative to the expected imbalance price, locking in volume at the intraday market pays; when the settlement price is expected to be higher, holding back volume to settle as imbalance pays. The governing principle is to tilt the bid in proportion to confidence in this spread signal and to curtail output when negative settlement prices are expected. Empirically, signal quality grows monotonically from the day-ahead to the setpoint stage: at chance at the day-ahead stage, moderate accuracy at the intraday stage, and strong enough at the setpoint stage to support a discrete fee-band classifier.