<p>Nowadays, with the growth of the population, the electricity demand is also rising, putting high pressure on the traditional grid. Microgrids have become an effective solution. This paper addresses a necessary challenge that directly impacts the assessment of microgrids With the rise of microgrids, renewable energy sources are being integrated and utilized in a more structured manner, alongside energy storage systems and residential power grids. This paper addresses a necessary challenge that directly impacts the assessment of microgrid efficiency: forecasting solar irradiance energy. The study utilizes a Reinforcement Learning algorithm, specifically an Actor-Critic model with a hybrid architecture. This approach combines the feature extraction capabilities of Convolutional Neural Networks with the time-series handling efficiency of Long Short-Term Memory networks to evaluate performance. Solar irradiance forecasting is conducted for two different scenarios: short-term forecasting for 1-hour ahead with 5-minute, 10-minute, and 20-minute resolutions; and long-term forecasting for 1-day ahead with 1-hour resolution. The long-term forecasting incorporates weather forecast datasets, enabling a comparative analysis. The Actor-Critic algorithm shows enhancements over alternative models compared to other approaches with a MAE of 36.22 <InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="40866_2025_283_Article_IEq1.gif" Format="GIF" Height="20" Rendition="HTML" Resolution="72" Type="Linedraw" Width="47" /> </InlineMediaObject> <EquationSource Format="TEX">\(W{/}m^2\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>W</mi> <mo stretchy="false">/</mo> <msup> <mi>m</mi> <mn>2</mn> </msup> </mrow> </math></EquationSource> </InlineEquation>, RMSE of 61.01 <InlineEquation ID="IEq2"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="40866_2025_283_Article_IEq1.gif" Format="GIF" Height="20" Rendition="HTML" Resolution="72" Type="Linedraw" Width="47" /> </InlineMediaObject> <EquationSource Format="TEX">\(W{/}m^2\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>W</mi> <mo stretchy="false">/</mo> <msup> <mi>m</mi> <mn>2</mn> </msup> </mrow> </math></EquationSource> </InlineEquation>, and MAPE of 11.92% for short-term forecasting, and a MAPE of 15.34% for long-term forecasting when integrating forecasted data. The irradiance forecast results for the upcoming day lay the foundation for optimal scheduling of the microgrid; accurate forecasting will support the control and management of energy generation sources.</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

A Reinforcement Learning-Based Framework for Short-Term and Long-Term Solar Irradiance Forecasting in Microgrid

  • D.T Bui,
  • D.N Nguyen,
  • T.L.H Pham

摘要

Nowadays, with the growth of the population, the electricity demand is also rising, putting high pressure on the traditional grid. Microgrids have become an effective solution. This paper addresses a necessary challenge that directly impacts the assessment of microgrids With the rise of microgrids, renewable energy sources are being integrated and utilized in a more structured manner, alongside energy storage systems and residential power grids. This paper addresses a necessary challenge that directly impacts the assessment of microgrid efficiency: forecasting solar irradiance energy. The study utilizes a Reinforcement Learning algorithm, specifically an Actor-Critic model with a hybrid architecture. This approach combines the feature extraction capabilities of Convolutional Neural Networks with the time-series handling efficiency of Long Short-Term Memory networks to evaluate performance. Solar irradiance forecasting is conducted for two different scenarios: short-term forecasting for 1-hour ahead with 5-minute, 10-minute, and 20-minute resolutions; and long-term forecasting for 1-day ahead with 1-hour resolution. The long-term forecasting incorporates weather forecast datasets, enabling a comparative analysis. The Actor-Critic algorithm shows enhancements over alternative models compared to other approaches with a MAE of 36.22 \(W{/}m^2\) W / m 2 , RMSE of 61.01 \(W{/}m^2\) W / m 2 , and MAPE of 11.92% for short-term forecasting, and a MAPE of 15.34% for long-term forecasting when integrating forecasted data. The irradiance forecast results for the upcoming day lay the foundation for optimal scheduling of the microgrid; accurate forecasting will support the control and management of energy generation sources.