Scholarlink

As solar photovoltaic (PV) power gains prominence in the global energy scenerio due to distributed energy resources and net zero targets, the intermittent nature of solar energy poses challenges for reliable short-term forecasting. This research focuses on advancing the accuracy of short-term solar PV power generation forecasts by leveraging the capabilities of deep neural networks (DNNs). We explore and compare the performance of specific DNN architectures, including Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Bidirectional LSTM (Bi-LSTM). Using a dataset sourced from UK Power Networks for July 2021, the proposed models undergo rigorous training and validation. Results demonstrate the effectiveness of the Bi-LSTM model, showcasing superior forecasting metrics with a Mean Absolute Error (MAE) of 0.22, Mean Squared Error (MSE) of 0.11, Root Mean Squared Error (RMSE) of 0.34, and \(R^{2}\) value of 0.84. This research contributes to optimizing short-term solar PV forecasting by using the capabilities of deep neural networks to improve solar energy integration into the developing energy paradigm.

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

Deep Learning Approach for Dynamic Short-Term Solar PV Power Generation Forecasting

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Deepesh Kumar,
  • Hemant Kumar Meena,
  • Prerna Jain

摘要

As solar photovoltaic (PV) power gains prominence in the global energy scenerio due to distributed energy resources and net zero targets, the intermittent nature of solar energy poses challenges for reliable short-term forecasting. This research focuses on advancing the accuracy of short-term solar PV power generation forecasts by leveraging the capabilities of deep neural networks (DNNs). We explore and compare the performance of specific DNN architectures, including Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Bidirectional LSTM (Bi-LSTM). Using a dataset sourced from UK Power Networks for July 2021, the proposed models undergo rigorous training and validation. Results demonstrate the effectiveness of the Bi-LSTM model, showcasing superior forecasting metrics with a Mean Absolute Error (MAE) of 0.22, Mean Squared Error (MSE) of 0.11, Root Mean Squared Error (RMSE) of 0.34, and \(R^{2}\) value of 0.84. This research contributes to optimizing short-term solar PV forecasting by using the capabilities of deep neural networks to improve solar energy integration into the developing energy paradigm.