Enhancing stochastic optimization of wind-PV-storage systems: A scenario reconstruction approach with source-load matching and temporal characteristics

The inherent stochasticity of renewable energy poses challenges to grid stability, prompting the widespread adoption of wind-photovoltaic-thermal-storage integrated systems (WPTS). While scenario-based stochastic programming (SBP), which utilizes scenario generation and reduction (often implemented via distance-based clustering such as K-means), is prevalent for planning such systems, our analysis reveals critical limitations. Through eight experimental on-grid cases based on meteorological data from 2009 to 2018 and validated against actual conditions from 2019 to 2023, SBP methods show significant deviations in Annual Cash Flow (Ct, a net present value component, 2.02%–23.08%), Renewable Energy Curtailment Rate (RECR, 74.83%–99.01%), and Power Tracking Deviation (RMSE, 2.15%–14.45%). These deviations fundamentally stem from the failure to preserve two critical data characteristics during conventional clustering-based scenario reduction: (1) source-load matching characteristics (i.e., the precise, point-in-time relationship between renewable generation and scheduled grid dispatch) and (2) temporal characteristics (i.e., the chronological sequence and evolution of these source-load relationships). To overcome these limitations, this paper proposes a scenario reconstruction approach. Developed through a targeted analysis of these limitations, this approach conducts hourly point clustering instead of 24-h line clustering and then reconstructs scenarios by emulating temporal patterns to preserve chronological features. Validations under the unified capacity planning scheme demonstrate the effectiveness of the improved framework. Under the improved framework, the relative deviations of Ct, RECR, and RMSE from their 2019–2023 actual mean values are reduced to 0.40%, 6.14%, and 0.86%, respectively (all within the observed fluctuation ranges). Compared with the conventional framework, these deviations magnitudes are further reduced by 11.34%, 79.89%, and 10.48%, respectively. More notably, even when compared to the typical meteorological year (TMY) deterministic case, which represents the best-performing baseline among all cases, the proposed approach still achieves further reductions in deviation of 4.46% for Ct and 1.94% for RMSE. The results demonstrate the enhanced rationality and applicability of the proposed stochastic optimization approach.