A Novel Time Diffusion Model for Industrial Time Series Data Generation

Given the intricate temporal correlations inherent in most process industry systems, temporal dynamic modeling methods based on time series data have demonstrated superior performance in the area of industrial data-driven modeling. However, due to constraints in data acquisition and maintenance methods, obtaining a substantial volume of high-quality and intact time series data from process industries is often challenging. To address this challenge, AI-Generated Content technologies (AIGC) can be employed to produce high-quality virtual data and ensure data sufficiency. This paper proposes a novel Time Diffusion Model (TDM) for the generation of industrial time series data. TDM introduces diffusion models to better capture the temporal correlations and data distributions of industrial time series data, enabling the generation of high-quality virtual time series data. This capability extends its applicability to time series data imputation and time series data augmentation tasks. The proposed TDM model is applied to the practical primary reformer process case. The experimental results substantiate that TDM can generate industrial time series data of high quality and effectively enhance the performance of industrial data-driven tasks.