Scholarlink

<p>The solar photovoltaic (PV) power has evolved to provide a concrete solution to high power demand, and green energy requirement. Erratic and fluctuating nature of solar energy, complicate the operation for the electrical grid to manage the power flow, when integrated to main grid. Furthermore, one of the prominent challenges of solar power is its variable availability resulting in low reliability. To better deal with it, a solid planning with least prediction error in solar irradiation/ solar power is extremely desirable. As a solution, a hybrid short-term prediction model, i.e., Wavelet Transform-based Gated-Recurrent Unit network (WTGRU) is presented in this paper. Four different families are compared to select the best mother wavelet family<i>, i.e.,</i> Daubechies family based decomposed solar PV power time series data is fed into the GRU, with dropout regularization model, along with important meteorological features. Utilizing Wavelet Reconstruction of the GRU output, the solar PV power is computed. The GRU, Long Short-Term Memory (LSTM), and ARIMA models are also considered for performance comparison. On the basis of normalized mean absolute error (nMAE) values and normalized root mean square error (nRMSE) values, it is concluded that the proposed hybrid model outperforms these models, indicating that it can greatly improve forecasting ability and reliability. Additionally, the model is tested and validated with longer duration solar PV power forecasting for 15&#xa0;days and 30&#xa0;days.</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

Wavelet Transform Based Gated-Recurrent Unit Deep Learning Approach for Power Output of Solar Photovoltaic System Forecasting

  • 期刊论文

  • 发表日期：

    2025年3月1日

查看全文
  • Prashant Singh,
  • Navneet Kumar Singh,
  • Asheesh Kumar Singh

摘要

The solar photovoltaic (PV) power has evolved to provide a concrete solution to high power demand, and green energy requirement. Erratic and fluctuating nature of solar energy, complicate the operation for the electrical grid to manage the power flow, when integrated to main grid. Furthermore, one of the prominent challenges of solar power is its variable availability resulting in low reliability. To better deal with it, a solid planning with least prediction error in solar irradiation/ solar power is extremely desirable. As a solution, a hybrid short-term prediction model, i.e., Wavelet Transform-based Gated-Recurrent Unit network (WTGRU) is presented in this paper. Four different families are compared to select the best mother wavelet family, i.e., Daubechies family based decomposed solar PV power time series data is fed into the GRU, with dropout regularization model, along with important meteorological features. Utilizing Wavelet Reconstruction of the GRU output, the solar PV power is computed. The GRU, Long Short-Term Memory (LSTM), and ARIMA models are also considered for performance comparison. On the basis of normalized mean absolute error (nMAE) values and normalized root mean square error (nRMSE) values, it is concluded that the proposed hybrid model outperforms these models, indicating that it can greatly improve forecasting ability and reliability. Additionally, the model is tested and validated with longer duration solar PV power forecasting for 15 days and 30 days.