Artificial Intelligence in Operation of Digitalized Energy Systems: Missing Data Imputation in the Modern Power System

Integrating renewable energy sources into the power grid presents complex challenges that require advanced operation and control in modern power systems. Phasor measurement units (PMUs) and wide-area measurement systems (WAMS) offer significant potential for monitoring power system operations and improving their stability. These systems are renowned for their impressive synchronization, speed, and precision. However, real-world applications may face hurdles like data loss due to communication congestion, hardware failures, or transmission delays. Consequently, this poses a substantial obstacle to power system functions, such as assessing security and stability. Traditionally, power system operation inputs, like PMU measurements, have been assumed to be constantly available for training and application. Practical issues like PMU malfunctions and communication congestion can disrupt this assumption. Furthermore, power system operation heavily relies on the observability of PMUs and the system’s topologies. If either of these factors changes or we cannot predict the exact value of missing data, the model’s effectiveness may decrease, thereby necessitating updates. This chapter aims to provide a comprehensive overview of various techniques for recovering missing time series data. Our review categorizes time series data imputation methods into four groups: traditional time series imputation methods, large-scale imputation algorithms, deep learning-based imputation methods, and generative adversarial networks. We will emphasize the literature on data-driven approaches and methods for recovering missing data.