Scholarlink

<p>Accurate forecasting of solar irradiance (SI) is crucial for the optimal functioning of solar power plants, given the inherently stochastic nature of SI and its dependency on meteorological variables. In this paper, we propose a novel hybrid multi-head forecasting model designed for very short-term (5 min) prediction. The model integrates a convolutional neural network (CNN), a high-precision long short-term memory (LSTM), and two gated recurrent units (GRU) layers. Each head independently trains meteorological input variables while preserving SI as the output. The fusion of outputs from individual heads, facilitated by fully connected layers, enhances the predictive precision of the model. The study not only assesses the model’s efficacy against advanced forecasting models, including LSTM, CNN, CNN-LSTM, and CNN-GRU, using 5-minute interval datasets but also evaluates its robustness against cyber-attacks. This includes false data injection (FDI) attacks, providing insights into the model’s resilience in the face of potential threats. Additionally, performance evaluations extend to datasets with 30 and 60-minute data samples. Notably, the proposed multi-head model outperforms the aforementioned models in terms of accuracy metrics, except in univariate SI forecasting. The proposed model achieves high predictive accuracy, recording a Root Mean Square Error (RMSE) of 26.32, Mean Absolute Error (MAE) of 13.09, and a coefficient of determination (R<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(^{2}\)</EquationSource> <EquationSource Format="MATHML"><math> <mmultiscripts> <mrow /> <mrow /> <mn>2</mn> </mmultiscripts> </math></EquationSource> </InlineEquation>) of 0.97 in the multivariate forecasting.</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 Solar Irradiance Forecasting Accuracy with a Multi-Head Weather-Aware Hybrid Model

  • 期刊论文

  • 发表日期：

    2025年10月1日

查看全文
  • Md. Omer Faruque,
  • Khadija Akter,
  • S. M. Mahfuz Alam,
  • Md. Alamgir Hossain,
  • Md. Rashidul Islam,
  • Md Shafiul Alam,
  • Md. Mahmudur Rahman

摘要

Accurate forecasting of solar irradiance (SI) is crucial for the optimal functioning of solar power plants, given the inherently stochastic nature of SI and its dependency on meteorological variables. In this paper, we propose a novel hybrid multi-head forecasting model designed for very short-term (5 min) prediction. The model integrates a convolutional neural network (CNN), a high-precision long short-term memory (LSTM), and two gated recurrent units (GRU) layers. Each head independently trains meteorological input variables while preserving SI as the output. The fusion of outputs from individual heads, facilitated by fully connected layers, enhances the predictive precision of the model. The study not only assesses the model’s efficacy against advanced forecasting models, including LSTM, CNN, CNN-LSTM, and CNN-GRU, using 5-minute interval datasets but also evaluates its robustness against cyber-attacks. This includes false data injection (FDI) attacks, providing insights into the model’s resilience in the face of potential threats. Additionally, performance evaluations extend to datasets with 30 and 60-minute data samples. Notably, the proposed multi-head model outperforms the aforementioned models in terms of accuracy metrics, except in univariate SI forecasting. The proposed model achieves high predictive accuracy, recording a Root Mean Square Error (RMSE) of 26.32, Mean Absolute Error (MAE) of 13.09, and a coefficient of determination (R \(^{2}\) 2 ) of 0.97 in the multivariate forecasting.