Scholarlink

<p>The intermittency of photovoltaic (PV) power output presents a major challenge for its integration into large-scale power grids. This study compares two machine learning models, CatBoost and HGBoost and develops hybrid versions optimized with FOA, GBO, and GWO to improve forecasting accuracy. Results show that while CatBoost achieves better standalone prediction, the HGBoost-GBO hybrid model performs best overall, with the lowest RMSE (54.28) and highest R² (0.969), compared to the CatBoost-FOA model with the highest RMSE and lowest R² (0.962). A bi-objective optimization problem is also introduced to maximize total energy output and minimize daily variance, with solutions generated using NSGA-II and AGE-MOEA. The top solution obtained with AGE-MOEA achieved a TOPSIS score of 0.946 and a VIKOR Q-value of 0.008, while NSGA-II’s best solution scored 0.93 and 0.076, respectively. The integration of multi-objective optimization (NSGA-II, AGE-MOEA) with multi-criteria decision-making methods (TOPSIS, VIKOR) provides a novel practical framework for grid integration. Additionally, the use of parallel coordinate plots for visualizing trade-offs among decision variables represents a rare but valuable contribution in PV optimization studies, enhancing transparency and interpretability of results. NSGA-II yields more stable results, while AGE-MOEA offers broader energy output ranges. TOPSIS and VIKOR are applied for solution ranking, and parallel coordinate plots are used for visualization. The proposed framework demonstrates strong potential for enhancing PV forecasting, stability, and grid integration.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing photovoltaic power output optimization: hybrid machine learning models for multi-objective and multi criteria decision making and sustainable energy integration

  • 期刊论文

  • 发表日期：

    2025年10月3日

查看全文
  • Chengfu Sun,
  • Hua Yang

摘要

The intermittency of photovoltaic (PV) power output presents a major challenge for its integration into large-scale power grids. This study compares two machine learning models, CatBoost and HGBoost and develops hybrid versions optimized with FOA, GBO, and GWO to improve forecasting accuracy. Results show that while CatBoost achieves better standalone prediction, the HGBoost-GBO hybrid model performs best overall, with the lowest RMSE (54.28) and highest R² (0.969), compared to the CatBoost-FOA model with the highest RMSE and lowest R² (0.962). A bi-objective optimization problem is also introduced to maximize total energy output and minimize daily variance, with solutions generated using NSGA-II and AGE-MOEA. The top solution obtained with AGE-MOEA achieved a TOPSIS score of 0.946 and a VIKOR Q-value of 0.008, while NSGA-II’s best solution scored 0.93 and 0.076, respectively. The integration of multi-objective optimization (NSGA-II, AGE-MOEA) with multi-criteria decision-making methods (TOPSIS, VIKOR) provides a novel practical framework for grid integration. Additionally, the use of parallel coordinate plots for visualizing trade-offs among decision variables represents a rare but valuable contribution in PV optimization studies, enhancing transparency and interpretability of results. NSGA-II yields more stable results, while AGE-MOEA offers broader energy output ranges. TOPSIS and VIKOR are applied for solution ranking, and parallel coordinate plots are used for visualization. The proposed framework demonstrates strong potential for enhancing PV forecasting, stability, and grid integration.