Energy management of pumped thermal electricity storage: a framework integrating a rule-based expert system with robust data-driven control

Pumped thermal electricity storage (PTES) is a promising grid-level energy storage technology for mitigating renewable power curtailment. However, its energy management remains challenging. PTES dynamics are nonlinear, mode-dependent, and difficult to model accurately over the full operating range. In renewable-rich microgrids, PTES operation further involves discrete decisions on charging, discharging, dispatchable generation, and grid power exchange. Minimum power limits make this discrete structure unavoidable. Embedding these discrete decisions in data-driven predictive control leads to a mixed-integer problem whose computational burden grows with the number of binary variables. This paper proposes a hierarchical energy management framework, RBES-ZPC, that integrates a forecast-informed rule-based expert system (RBES) with data-driven zonotopic predictive control (ZPC). The framework separates discrete decisions from continuous power allocation. The RBES resolves the discrete decisions and directly determines control actions away from constraint-critical regions. These actions are implemented only after passing a safety verification. Near such regions or if the verification fails, the ZPC is activated and solves a continuous optimization problem. Input–output data are collected separately for each PTES operating mode. The resulting mode-specific formulations reduce conservatism compared with a single-model design and preserve conditional robust closed-loop constraint satisfaction under bounded process and measurement noise. Case studies show that PTES integration under the RBES-ZPC effectively reduces renewable power curtailment and net operating cost under the considered operating conditions. Compared with a mixed-integer ZPC, the RBES-ZPC reduces the mean total online computation time from 99.60 s to 26.00 s with comparable operating performance.