Advancing scenario generation in large-scale clean energy bases via enhanced hyperparameter optimization techniques

With the rapid expansion of large-scale clean energy bases in major energy-producing countries, high-quality scenario generation has become essential for effective energy management and intelligent scheduling. This study introduces a hyperparameter optimization method for a Least Squares Generative Adversarial Network (LSGAN) based on the PID-based Search Algorithm with Joint Opposite Selection (PSA-JOS), driven by a multi-layer fully connected perceptron. The optimization objective is defined as the Wasserstein distance between the original and generated scenarios. The PSA-JOS algorithm, developed through structural advancements in the original PSA framework, demonstrates superior performance, as validated through benchmark function tests. The average second-order Wasserstein distance serves as a quantitative metric to assess the distributional discrepancy between the original and generated scenarios, effectively reflecting the generation quality. Following hyperparameter optimization, the LSGAN exhibits enhanced performance in generating one-day scenarios for wind power, direct normal irradiation (DNI), and load power, achieving a substantial reduction in the average Wasserstein distance over multiple iterations. The optimized generative adversarial network not only provides reliable data support but also enhances decision-making capabilities for the future expansion and intelligent scheduling of clean energy bases. This study offers new insights into complex energy system modeling using generative adversarial networks and presents an effective approach for capturing multivariate temporal variations and generating realistic energy scenarios.