Optimization of coupled electric-thermal-ammonia energy systems with demand response using chaotic-swarm enhanced sparrow search

• Dual-stage stochastic model for electric-thermal-ammonia energy systems. • Bi-level model reduces cost by 5.8% and emissions by 31.5% vs. baseline system. • CS-SSA eliminates wind curtailment; converges 15–30% faster than SSA/PSO/GSA. • P2A2P system cuts total emissions by 20.3% in intra-day dispatch scenario. • Ammonia storage charges up to 17 MW, discharges 20 MW for load balance. As smart buildings evolve into key nodes of future low-carbon energy networks, managing their complex energy demands across electricity, heat, and chemical carriers presents a growing challenge. However, most existing studies focus on deterministic or single-stage dispatch models and rarely integrate P2A conversion and SOAFC-based power regeneration within a bi-level stochastic scheduling framework. This paper proposes a bi-level stochastic optimization framework for smart building energy management that integrates coupled electric–thermal–ammonia systems. The system architecture incorporates Power-to-Ammonia (P2A) conversion, Solid Oxide Ammonia Fuel Cells (SOAFCs), and multi-scale demand response mechanisms, enabling flexible energy storage, cross-domain energy conversion, and enhanced operational resilience. The model operates across two temporal layers, including day-ahead and intra-day scheduling, capturing both forecast uncertainty and real-time variability in renewable generation and building load profiles. To solve the resulting high-dimensional, non-convex problem, a Chaotic-Swarm Enhanced Sparrow Search Algorithm (CS-SSA) is developed, combining logistic map–based chaotic initialization, adaptive role dynamics, and penalty-based constraint handling. Applied to a regional smart building cluster integrated with wind power, the proposed method achieves a 5.8% reduction in operational cost, a 31.5% decrease in carbon emissions, and elimination of wind curtailment under the tested intra-day dispatch scenarios. Compared with benchmark algorithms, CS-SSA demonstrates improved empirical performance on the tested scheduling problem, achieving a 6.3% lower operating cost and reduced average computation time of 38.5 s (compared with 45.2 s for SSA over 30 runs).