Scholarlink

<p>Solar radiation forecasting plays a crucial role in mitigating climate change, ensuring stable and reliable utilization of solar energy and conducting feasibility studies for solar energy projects. Capturing temporal trends in non-stationary solar radiation data often requires multiple methodologies. In this context, this study proposes an Seasonal Trend Loess based hybrid model that combines the advantages of Seasonal Autoregressive Integrated Moving Average (SARIMA), which effectively explains linear relationships, and Long Short-Term Memory (LSTM), which excels in capturing nonlinear dependencies. The Loess smoothing technique, which is particularly effective for time series with high seasonal variability, enables the decomposition of the time series into seasonal, trend and residual components for separate modeling. Furthermore, to assess the model's stability across all periods and evaluate its predictive performance, tenfold cross-validation was applied to the time series. The study utilized hourly measured solar irradiation data for Afyonkarahisar, Turkey. When compared to standalone SARIMA and LSTM forecasting models, the proposed hybrid model achieved improvements of 37.84% in RMSE, 37.45% in NRMSE, 34.03% in MAE, and 4.42% in R<sup>2</sup> metrics.</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 novel approach to solar radiation forecasting using a SARIMA–LSTM hybrid model with seasonal trend loess decomposition

  • 期刊论文

  • 发表日期：

    2025年7月7日

查看全文
  • Feyza Nur Yeşil,
  • Tuba Nur Serttaş

摘要

Solar radiation forecasting plays a crucial role in mitigating climate change, ensuring stable and reliable utilization of solar energy and conducting feasibility studies for solar energy projects. Capturing temporal trends in non-stationary solar radiation data often requires multiple methodologies. In this context, this study proposes an Seasonal Trend Loess based hybrid model that combines the advantages of Seasonal Autoregressive Integrated Moving Average (SARIMA), which effectively explains linear relationships, and Long Short-Term Memory (LSTM), which excels in capturing nonlinear dependencies. The Loess smoothing technique, which is particularly effective for time series with high seasonal variability, enables the decomposition of the time series into seasonal, trend and residual components for separate modeling. Furthermore, to assess the model's stability across all periods and evaluate its predictive performance, tenfold cross-validation was applied to the time series. The study utilized hourly measured solar irradiation data for Afyonkarahisar, Turkey. When compared to standalone SARIMA and LSTM forecasting models, the proposed hybrid model achieved improvements of 37.84% in RMSE, 37.45% in NRMSE, 34.03% in MAE, and 4.42% in R2 metrics.