This paper studies the time series forecasting problem from a whole new
perspective. In the existing SOTA time-series representation learning methods,
the forecasting models take a sequence of numerical values as input and yield
numerical values as output. The existing SOTA models are largely based on
Transformer architecture, modified with multiple encoding mechanisms to
incorporate the context and semantics around the historical data. In this
paper, we approach representation learning of time-series from the paradigm of
prompt-based natural language modeling. Inspired by the successes of
pre-trained language foundation models, we pose a question about whether these
models can also be adapted to solve time-series forecasting. Thus, we propose a
new forecasting paradigm: prompt-based time series forecasting (PromptCast). In
this novel task, the numerical input and output are transformed into prompts.
We frame the forecasting task in a sentence-to-sentence manner which makes it
possible to directly apply language models for forecasting purposes. To support
and facilitate the research of this task, we also present a large-scale dataset
(PISA) that includes three real-world forecasting scenarios. We evaluate
different SOTA numerical-based forecasting methods and language generation
models such as Bart. The benchmark results with single- and multi-step
forecasting settings demonstrate that the proposed prompt-based time series
forecasting with language generation models is a promising research direction.
In addition, in comparison to conventional numerical-based forecasting,
PromptCast shows a much better generalization ability under the zero-shot
setting. We believe that the proposed PromptCast task as well as our PISA
dataset could provide novel insights and further lead to new research
directions in the domain of time-series representation learning and
forecasting.

本文探讨了基于提示的自然语言建模的范式下的时间序列表示学习问题，提出了基于提示的时间序列预测 (PromptCast) 模型，并介绍了一个大规模数据集 (PISA)，结果表明这种基于语言生成模型的提示方法是一种有前途的研究方向。