Comparative Evaluation of Multi-Timescale Scheduling Strategies for a CSP Plant with Thermal Energy Storage

Abstract

Forecast errors in direct normal irradiance (DNI) and load demand can reduce the economic performance of concentrated solar power (CSP) plants with thermal energy storage (TES). This study compares a day-ahead-only strategy (DA-only), rolling particle swarm optimization (DA + RollPSO), and model predictive control (DA + MPC) in a common CSP-TES-grid framework. The analysis covers 12 independent monthly representative 24 h cases with synthetically generated DNI and load forecast errors. All strategies use the same day-ahead baseline, system parameters, physical constraints, and error realizations. Relative to DA-only, DA + RollPSO reduced total operating cost and grid purchases by 1.48% and 1.40%, while DA + MPC achieved reductions of 1.47% and 1.37%. The archived post-correction main comparison used tracking-augmented objectives for both intraday methods with identical normalized weights. DA + RollPSO achieved a TES stored-energy RMSE of 16.22 MWhth, whereas DA + MPC produced 18.09 MWhth with a shorter repeated mean single-step solution time (3.219 versus 43.850 ms). Both intraday strategies also reduced cost relative to DA-only in all 20 documented paired Monte Carlo trials. These results indicate a case-specific trade-off between TES trajectory tracking and computational burden, not universal superiority of either controller.