Load Forecasting Model Trading: A Cost-Oriented and Auction-Based Approach

Data sharing is essential for accurate load forecasting and efficient energy management, yet data exchange remains severely constrained by a lack of effective economic incentives. Data markets have emerged as a potential solution by incentivizing the exchange of data and forecasting resources among stakeholders. Existing data market mechanisms, often designed around trading raw data or forecast outputs, face multiple barriers—such as privacy concerns, valuation misaligned with real operational benefits, unilateral pricing, computationally intensive allocation, and inflexible asset adaptation—that limit their practicality for real-world energy applications. The aim of this paper is to address these challenges by proposing a novel market framework that treats pre-trained forecasting models as tradable assets, thereby fundamentally redefining the data market paradigm. Specifically, a cost-oriented evaluation approach that links model quality to downstream operational costs is first established as the foundation throughout the entire market process. Subsequently, we propose a bilateral iterative model auction mechanism to enable efficient transactions between the buyer and sellers while maximizing social welfare. Furthermore, we propose a model adaptation strategy, including model fine-tuning and ensembling, for the buyer to enhance the applicability of purchased models to his decision-making problem. Case studies in building energy management based on public datasets demonstrate that our approach converges to the socially optimal solution, allowing all participants to benefit: sellers are appropriately compensated for providing high-quality models, and buyers achieve significant operational cost reductions through the utilization of traded models.