Decision-Focused Fine-Tuning of Time Series Foundation Models for Dispatchable Feeder Optimization

2025-03-03

AI Summary

In power distribution systems for building-scale applications, scarce data impedes high-value, task-specific forecasting required for optimizing dispatchable feeders. Method: This paper proposes a decision-oriented fine-tuning paradigm for temporal foundation models, embedding downstream optimization objectives directly into an end-to-end differentiable training pipeline. Leveraging Moirai as the base model, we integrate parameter-efficient fine-tuning (PEFT) with a decision-focused loss function and a differentiable optimization layer to enable gradient propagation and instance-specific prediction. Results: On dispatchable feeder optimization, our method reduces average daily total cost by 9.45% over prediction-centric baselines, significantly improving operational economics. Our core contribution is the first systematic integration of decision-focused learning into temporal foundation model fine-tuning—enabling customized, low-sample, high-decision-utility forecasting explicitly aligned with optimization goals.

Technology Category

• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Search and Optimization: Learning to Search
• Machine Learning: Optimization

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Time series foundation models provide a universal solution for generating forecasts to support optimization problems in energy systems. Those foundation models are typically trained in a prediction-focused manner to maximize forecast quality. In contrast, decision-focused learning directly improves the resulting value of the forecast in downstream optimization rather than merely maximizing forecasting quality. The practical integration of forecast values into forecasting models is challenging, particularly when addressing complex applications with diverse instances, such as buildings. This becomes even more complicated when instances possess specific characteristics that require instance-specific, tailored predictions to increase the forecast value. To tackle this challenge, we use decision-focused fine-tuning within time series foundation models to offer a scalable and efficient solution for decision-focused learning applied to the dispatchable feeder optimization problem. To obtain more robust predictions for scarce building data, we use Moirai as a state-of-the-art foundation model, which offers robust and generalized results with few-shot parameter-efficient fine-tuning. Comparing the decision-focused fine-tuned Moirai with a state-of-the-art classical prediction-focused fine-tuning Morai, we observe an improvement of 9.45% in average total daily costs.

Problem

• Improves forecast value in energy system optimization
• Addresses complex applications with diverse instances
• Enhances decision-focused learning for feeder optimization

Innovation

• Decision-focused fine-tuning enhances optimization outcomes
• Moirai model enables robust predictions with scarce data
• Improves daily costs by 9.45% over traditional methods