Advancing hybrid system optimization via value-oriented forecasting: An application in hydro-wind-photovoltaic operation
Year: 2025
Type: article
Abstract: • A novel closed-loop forecast-and-optimize framework for day-ahead generation and reserve scheduling is successfully developed. • Employ orthogonal experimental design to reduce exhaustive runs while ensuring a value-oriented optimal scheduling plan. • Propose knowledge-guided meta -learning modules to enable direct scheduling without repeated closed-loop traversal. • Design five experimental models across different seasons to examine the effectiveness and efficiency. Day-ahead generation and reserve scheduling generally follows a forecast-then-optimize process, but reliance on static prediction intervals renders this open-loop task ill-suited for high-uncertainty power systems. This study proposes a closed-loop forecast-and-optimize framework with knowledge-guided meta -learning modules, embedding orthogonal experimental design to transform static quantile intervals into dynamic, time-varying intervals. By iteratively aligning prediction intervals with scheduling outcomes through value-oriented feedback, the framework ensures an economically optimal day-ahead hydro-wind-photovoltaic generation and reserve scheduling plan. Furthermore, knowledge-guided meta -learning modules, complementing the orthogonal closed-loop framework, leverage prior scheduling knowledge and a feature-mapping network to transform repeated iterative traversal into efficient direct mapping, thereby accelerating closed-loop operations while preserving accuracy and value orientation. The closed-loop process is applied and verified in the Lianghekou hydro-wind-photovoltaic integrated energy base case study. Across successive iterations, the solution set becomes increasingly concentrated on higher net complementary profits with fewer violations, ensuring convergence toward balanced solutions that are both economically efficient and operationally reliable. Meanwhile, the closed-loop framework with knowledge-guided meta -learning modules achieves profitability nearly identical to the iterative orthogonal closed-loop framework, while reducing average runtime by 98.18%. Results indicate that proposed framework adapts well to hydropower operations under uncertainty, enabling safer, more economical, and flexible scheduling in high-uncertainty renewable systems.
Language: English
Subfield: Electrical and Electronic Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $3,170
OpenAlex ID: W7117586668