Scholarlink

<p>Eskom, South Africa’s main utility company, faces significant challenges in meeting the high energy demand of a growing population and increasing industrialization, largely due to aging coal power generation plants. Recent investments in renewable energy resources to accelerate the transition away from fossil fuels also encounter significant challenges, particularly due to the impacts of local weather variability on renewable energy output. To help improve Eskom’s grid management under weather uncertainty, we developed machine learning and Deep learning models to predict the short-term impacts of weather variability on Eskom’s renewable-to-grid integration at an hourly timescale, using data from ERA5 and Eskom. A comparative analysis between the Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Artificial Neural Network (ANN) shows the superiority of LSTM in predicting renewable energy generation under weather uncertainty, as they can capture autocorrelation and temporal patterns more effectively. We also investigate the impact of seasonality on Eskom’s renewable generation capacity. Seasonal analysis reveals a complementary relationship between wind and solar: when wind power output is low, solar output is high, and vice versa, which improves grid stability when both sources are combined. Future projections using CMIP6 under two emission scenarios (SSP2-4.5 and SSP5-8.5) reveal that, under moderate emission scenarios, we expect a strong increase in both solar radiation and wind speed compared to high emission scenarios, suggesting renewable energy resources are equally impacted by anthropogenic climate change. This work is very important as it can help Eskom to better manage its grid, avoid unforeseen power failures and load shedding, and support South Africa’s transition to a low-carbon energy future.</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 South Africa’s power grid management with AI: deep learning models for short-term forecasting of Eskom’s renewable generation under weather uncertainty

  • 期刊论文

  • 发表日期：

    2026年3月31日

查看全文
  • Gift Katlego Xelibokwe,
  • Nkongho Ayuketang Arreyndip,
  • Joseph Ebobenow,
  • David Afungchui

摘要

Eskom, South Africa’s main utility company, faces significant challenges in meeting the high energy demand of a growing population and increasing industrialization, largely due to aging coal power generation plants. Recent investments in renewable energy resources to accelerate the transition away from fossil fuels also encounter significant challenges, particularly due to the impacts of local weather variability on renewable energy output. To help improve Eskom’s grid management under weather uncertainty, we developed machine learning and Deep learning models to predict the short-term impacts of weather variability on Eskom’s renewable-to-grid integration at an hourly timescale, using data from ERA5 and Eskom. A comparative analysis between the Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Artificial Neural Network (ANN) shows the superiority of LSTM in predicting renewable energy generation under weather uncertainty, as they can capture autocorrelation and temporal patterns more effectively. We also investigate the impact of seasonality on Eskom’s renewable generation capacity. Seasonal analysis reveals a complementary relationship between wind and solar: when wind power output is low, solar output is high, and vice versa, which improves grid stability when both sources are combined. Future projections using CMIP6 under two emission scenarios (SSP2-4.5 and SSP5-8.5) reveal that, under moderate emission scenarios, we expect a strong increase in both solar radiation and wind speed compared to high emission scenarios, suggesting renewable energy resources are equally impacted by anthropogenic climate change. This work is very important as it can help Eskom to better manage its grid, avoid unforeseen power failures and load shedding, and support South Africa’s transition to a low-carbon energy future.