Scholarlink

<p>Accurate solar irradiance forecasting is essential for grid stability, photovoltaic integration, and energy management. This study presents a systematic literature review of artificial intelligence approaches for solar irradiance forecasting published between 2015 and 2024, following the Kitchenham protocol and PRISMA guidelines. A total of 80 peer-reviewed studies were analyzed through quantitative and qualitative assessments. Unlike previous reviews that mainly provide descriptive summaries, this study introduces a triadic analytical framework linking forecasting horizon, input data, and model architecture. Chi-square testing, Cramér’s V association analysis, Multiple Correspondence Analysis (MCA), and K-Modes clustering were employed to identify statistical associations, latent methodological structures, and dominant forecasting paradigms. The results reveal a progressive transition from traditional statistical and machine learning approaches toward deep learning and hybrid frameworks, with hybrid models representing more than 42% of recent studies. Short-Term forecasting is the most investigated horizon (39% of publications), while Global Horizontal Irradiance (GHI), temperature, relative humidity, and wind speed are the most frequently used input variables. Wavelet Transform was the most frequently adopted preprocessing technique, while Minimum Redundancy Maximum Relevance (MRMR) and metaheuristic optimization algorithms were commonly used for feature selection and parameter optimization. RMSE and R<sup>2</sup> were the evaluation metrics reported most frequently across the reviewed studies. The analyses indicate that methodological choices are influenced by the combined effects of forecasting horizon, input data characteristics, and model architecture rather than by model complexity alone. Overall, this review offers a quantitative synthesis of recent advances in AI-based solar irradiance forecasting and introduces a triadic analytical framework that facilitates the interpretation of methodological trends and supports future investigations.</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

A Triadic Analytical Framework for Solar Irradiance Forecasting: A Systematic Quantitative Review of AI-Based Methods

  • 期刊论文

  • 发表日期：

    2026年7月20日

查看全文
  • Saida Ghiate,
  • Jamal Zahi

摘要

Accurate solar irradiance forecasting is essential for grid stability, photovoltaic integration, and energy management. This study presents a systematic literature review of artificial intelligence approaches for solar irradiance forecasting published between 2015 and 2024, following the Kitchenham protocol and PRISMA guidelines. A total of 80 peer-reviewed studies were analyzed through quantitative and qualitative assessments. Unlike previous reviews that mainly provide descriptive summaries, this study introduces a triadic analytical framework linking forecasting horizon, input data, and model architecture. Chi-square testing, Cramér’s V association analysis, Multiple Correspondence Analysis (MCA), and K-Modes clustering were employed to identify statistical associations, latent methodological structures, and dominant forecasting paradigms. The results reveal a progressive transition from traditional statistical and machine learning approaches toward deep learning and hybrid frameworks, with hybrid models representing more than 42% of recent studies. Short-Term forecasting is the most investigated horizon (39% of publications), while Global Horizontal Irradiance (GHI), temperature, relative humidity, and wind speed are the most frequently used input variables. Wavelet Transform was the most frequently adopted preprocessing technique, while Minimum Redundancy Maximum Relevance (MRMR) and metaheuristic optimization algorithms were commonly used for feature selection and parameter optimization. RMSE and R2 were the evaluation metrics reported most frequently across the reviewed studies. The analyses indicate that methodological choices are influenced by the combined effects of forecasting horizon, input data characteristics, and model architecture rather than by model complexity alone. Overall, this review offers a quantitative synthesis of recent advances in AI-based solar irradiance forecasting and introduces a triadic analytical framework that facilitates the interpretation of methodological trends and supports future investigations.