Machine learning-driven paradigm shift in thermochemical energy storage

Thermochemical energy storage (TCES) technology, which enables high-density energy storage and flexible dispatch through reversible chemical reactions, is a critical solution to address the spatiotemporal mismatch of renewable energy. However, TCES technology has been difficult to widely diffuse because of the difficulty in discovering TCES materials with even better properties and the difficulty of TCES systems in accurately predicting the dynamic heat load demand of users in smart energy systems. Traditional research methods relying on experimental trial and error and simplified numerical simulations face challenges such as high costs, prolonged experimental cycles, and insufficient prediction accuracy. It is imperative that novel research tools be employed as soon as possible to address these two major issues. Machine Learning (ML), as a data-driven intelligent tool, transforms the exploration of causal relationships into correlational analyses, demonstrating exceptional potential in TCES for novel material screening and system optimization due to its advanced data processing capabilities, low hardware cost requirements, and high prediction accuracy. This perspective intensively describes the methodology for embedding ML algorithms in TCES, as well as describing the characteristics of the various algorithms. Furthermore, it outlines strategies to enhance the application of ML algorithms in TCES technologies, proposing pathways such as establishing standardized TCES databases through extensive data collection to broaden ML applicability, implementing weighted parameter optimization for input variables or combining physical theory in the input process of the underlying data to improve algorithmic stability, integrating transfer learning with ML frameworks to maintain accuracy under data scarcity, and training ML models with region-specific operational data to explore multi-dimensional and multi-objective optimization across the 4E criteria (Energy, Environment, Economic, Exergy) and interdisciplinary convergence. While current applications of ML in the TCES domain remain in their nascent stages, the technology holds significant potential to drive transformative advancements in future research and industrial deployment.