A review of deep reinforcement learning algorithms for grid services in grid-interactive efficient buildings

Integrating deep reinforcement learning (DRL) into building energy management systems offers an innovative method for improving grid stability and efficiency. Grid-interactive efficient buildings (GEBs) that utilize intelligent algorithms can dynamically optimize their energy usage and generation based on real-time grid conditions. This paper provides a comprehensive review of the applications of DRL in GEBs, focusing on its potential to autonomously manage energy consumption, balance renewable energy variability, and support grid services. The paper also examines the simulation tools available for evaluating GEB performance and highlight their integration with DRL frameworks. Despite the promising advancements, DRL applications encounter challenges such as algorithmic complexity, reliance on extensive data, and difficulties in integrating with existing infrastructures. The paper will also present comprehensive analyses that showcase current achievements while also investigate key challenges that must be addressed. Tackling these issues will allow DRL to realize its full potential in building energy management systems, contributing to a more sustainable and efficient energy future. • Focuses on DRL-based grid services in grid-interactive efficient buildings (GEBs). • Systematic 2015–2025 review with PRISMA-style screening and data extraction. • Maps core DRL algorithms to GEB grid services, control targets, and KPIs. • Synthesizes real-world DRL deployments in buildings. • Proposes constraint-aware DRL workflow with digital twins, safety guards, and TL.