Advances in Time Series Imputation and Energy Forecasting

The field of time series analysis is moving towards more robust and accurate methods for handling missing or irregularly sampled data. Researchers are developing innovative techniques that can effectively impute missing values and provide reliable frequency estimates, which is crucial for various applications such as energy forecasting and smart utility management. One of the key directions is the integration of spectral guidance into machine learning approaches, enabling the use of irregularly sampled data without requiring prior interpolation. This advancement has the potential to improve the accuracy of time series forecasting and other related tasks. Another important area of research is the development of robust imputation models that can handle diverse missing patterns in real-world datasets. This includes the creation of preprocessing frameworks that can bridge the gap between artificially masked training data and real missing patterns. Noteworthy papers in this area include LSCD, which introduces a novel score-based diffusion model for time series imputation conditioned on the entire signal spectrum, and DIM-SUM, which proposes a preprocessing framework for training robust imputation models that can handle diverse missing patterns. Additionally, the study on time-series surrogates from energy consumers and Industrial Energy Disaggregation with Digital Twin-generated Dataset are also making significant contributions to the field.

Sources

LSCD: Lomb-Scargle Conditioned Diffusion for Time series Imputation

DIM-SUM: Dynamic IMputation for Smart Utility Management

Time-series surrogates from energy consumers generated by machine learning approaches for long-term forecasting scenarios

Industrial Energy Disaggregation with Digital Twin-generated Dataset and Efficient Data Augmentation

Built with on top of