Scholarlink

<p>Accurate photovoltaic (PV) power forecasting is crucial for efficient energy management in microgrid systems, where predicting significant drops in energy production over two or three days is more important than predicting short-term energy output, especially when integrated with battery storage systems. In this study, we explore the benefits of using frequency-domain methods, such as the Discrete Wavelet Transform (DWT) and Short-Time Fourier Transform (STFT), in combination with Long Short-Term Memory (LSTM), a widely used time series prediction technique, and more recent approaches such as Transformers. To assess the robustness of the proposed methods, extensive tests and comparisons were preformed. This enabled the evaluation of prediction performance under varying weather conditions, including fog and cloudy scenarios. In addition, we evaluated the total energy produced across different time horizons. Performance assessment results for 1-hour horizon power forecasting revealed that the DWT-LSTM method achieved an average normalized root-mean-squared error (HH) of approximately 0.16, outperforming the classical CNN-LSTM method, which yielded an HH of more than 0.22. For longer-horizon forecasts, the LSTM-STFT-ANN method exhibited higher efficiency. Furthermore, "Vanilla" time series Transformers were tested individually and in conjunction with wavelet transform. For instance, for 3-day horizon, the HH indicator was recorded as 0.405 for the LSTM-STFT-ANN model, compared to around 0.410 for the Transformer-based model and 0.637 for the CNN-LSTM method.</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

Machine learning-based photovoltaic power and energy prediction in time–frequency domain

  • 期刊论文

  • 发表日期：

    2025年7月21日

查看全文
  • Abdelaziz El aouni,
  • Salah Eddine Naimi,
  • Yassine Ayat

摘要

Accurate photovoltaic (PV) power forecasting is crucial for efficient energy management in microgrid systems, where predicting significant drops in energy production over two or three days is more important than predicting short-term energy output, especially when integrated with battery storage systems. In this study, we explore the benefits of using frequency-domain methods, such as the Discrete Wavelet Transform (DWT) and Short-Time Fourier Transform (STFT), in combination with Long Short-Term Memory (LSTM), a widely used time series prediction technique, and more recent approaches such as Transformers. To assess the robustness of the proposed methods, extensive tests and comparisons were preformed. This enabled the evaluation of prediction performance under varying weather conditions, including fog and cloudy scenarios. In addition, we evaluated the total energy produced across different time horizons. Performance assessment results for 1-hour horizon power forecasting revealed that the DWT-LSTM method achieved an average normalized root-mean-squared error (HH) of approximately 0.16, outperforming the classical CNN-LSTM method, which yielded an HH of more than 0.22. For longer-horizon forecasts, the LSTM-STFT-ANN method exhibited higher efficiency. Furthermore, "Vanilla" time series Transformers were tested individually and in conjunction with wavelet transform. For instance, for 3-day horizon, the HH indicator was recorded as 0.405 for the LSTM-STFT-ANN model, compared to around 0.410 for the Transformer-based model and 0.637 for the CNN-LSTM method.