A market-rule-informed neural network for efficient imbalance electricity price forecasting

Abstract

Accurate and efficient imbalance electricity price forecasting is critical for industrial energy trading in balancing markets. Battery assets and automated trading systems require forecasts with limited computational resources and near-real-time speed. The imbalance price is formed through known market rules that aggregate various market signals into the final market outcome. For accurate price forecasting, this raises a methodological question: can future prices be sufficiently forecast from only lagged prices, or are raw market signals needed? This paper argues for the latter and proposes a market-rule-informed neural network (MRINN) that embeds market rules into the latent space. By embedding these rules, the proposed model preserves raw signal information while avoiding the need to relearn known market rules, leading to a compact model with faster training and inference. We further examine operational robustness under delayed or missing signals and characterize performance scaling across input length and forecasting horizon. Our case studies provide empirical evidence that the proposed MRINN improves forecasting accuracy by 8% compared with the best lagged-price-only baseline, while reducing the number of trainable parameters by around 90% relative to raw-signal neural baselines. The data and code are open-sourced at https://github.com/runyao-yu/MRINN .