Market-Information-Aware Gated-LoRA Framework Enhances Cross-Market Transferability for Electricity Price Forecasting
By Mr.Xu
Published:
Summary:This research proposes a Market-Information-Aware Gated-LoRA framework for transferring the Chronos-2 time-series foundation model to day-ahead electricity price forecasting. By constructing a Multi-Source Market Information (MSMI) interface and training a gated low-rank adapter (LoRA) that updates only 1% of the model parameters without requiring target-market labels, the framework achieves significant improvements in cross-market transferability. Experiments on four Chinese provincial day-ahea
Background and Challenges
Electricity price forecasting is crucial for market participants but faces challenges such as price volatility, market specificity, and close ties to anticipated system conditions. Existing supervised learning methods heavily rely on market-specific historical data, which limits their applicability in newly established or data-scarce markets.
Methodology and Innovation
The proposed Market-Information-Aware Gated-LoRA framework includes the following steps:
- Multi-Source Market Information (MSMI) Interface Construction: Aligns 7-day price context with pre-clearing supply-demand, reserve, maintenance, generator-capacity, and intertie variables.
- Source-Domain Gated Low-Rank Adapter (LoRA) Training: Updates only about 1% of the model parameters and uses reserve-tightness and operating-state signals to scale the frozen source adapter.
- Cross-Market Transferability Evaluation: Employs a leave-one-market-out protocol to ensure the model's generalization capabilities.
Experimental Results
Experiments on four Chinese provincial day-ahead spot markets show that the framework reduces the average MAE/RMSE by 6.24%/7.99% compared to market-information-aware zero-shot Chronos-2 and by 3.05%/3.52% compared to vanilla Source-LoRA. The results also indicate that the gain cannot be reproduced by a learned global scalar or by random gate initialization, while the additional improvement over Source-LoRA is limited. These findings validate the practical benefits of market-structured inputs and state-dependent gated LoRA for data-scarce electricity markets.
Industry Impact and Developer Recommendations
- Industry Impact: This research provides a practical migration path for electricity price forecasting in data-scarce markets, helping market participants improve prediction accuracy and decision-making.
- Developer Recommendations: Developers are encouraged to apply similar gated adapter methods in analogous data-scarce scenarios and to incorporate domain knowledge to build more effective multi-source information interfaces.
Conclusion
Market-structured inputs and state-dependent gated LoRA offer an effective migration method for cross-market electricity price forecasting, particularly advantageous in data-scarce situations.
References
— END —Source: ArXiv Machine Learning (cs.LG) (2026-08-13)
Tags: #Chronos-2 #Gated-LoRA #Electricity Price Forecasting #Cross-Market Transferability #Data-Scarce
Community Comments