Experimental Transformer System for Time Series Forecasting

Abstract

The goal of the project is to use and suit a transformer based foundation model such as Lag-Llama (https://arxiv.org/abs/2310.08278) to forecast energy prices. Lag-Llama is pretrained on a large corpus of diverse time series data from several domains, and demonstrates strong zero-shot generalization capabilities compared to a wide range for forecasting models. The forecasting task should focus on general one-step (t+1) and multi-step predictions (t+n). Forecasts that can predict extremes (maximum prices and minimum prices ) are especially important. Both univariate and multivariate forecasting should be addressed in accordance with findings documented in https://m unin.uit.no/handle/10037/32757?show=full&locale-attribute=en.