Scholarlink

<p>Accurate short-term net load forecasting is essential for reliable operation of renewable energy-integrated microgrids. This paper presents a parallel ensemble PSO-ANN framework comprising three independently optimized Artificial Neural Network (ANN) modules for solar power, wind power, and electrical load demand forecasting, whose outputs are fused to derive the net load. Particle Swarm Optimization (PSO) is employed specifically to optimize the weights and biases of each ANN prior to the supervised learning phase, replacing conventional gradient-based initialization and thereby improving convergence characteristics and avoiding local minima entrapment. The parallel ensemble structure is motivated by the distinct statistical properties of each generation and consumption component, enabling specialized feature learning for each sub-problem before integration. Experimental validation using chronologically partitioned real-world operational data demonstrates that the proposed framework achieves MAPE values of 5.3%, 8.7%, and 3.8% for solar, wind, and load forecasting, respectively, and an overall net load MAPE of 6.2%, outperforming conventional methods including ARIMA, standard backpropagation ANN, and Support Vector Regression, as well as benchmark deep learning models. The results confirm that the PSO-optimized parallel ensemble approach provides a robust and computationally tractable solution for net load forecasting under high renewable penetration.</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

Short-term load forecasting in renewable energy microgrids using PSO-optimized parallel artificial neural networks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月3日

查看全文
  • Jigneshkumar Desai,
  • Rakeshkumar A. Patel,
  • Nikunj Patel,
  • Meet Sureja

摘要

Accurate short-term net load forecasting is essential for reliable operation of renewable energy-integrated microgrids. This paper presents a parallel ensemble PSO-ANN framework comprising three independently optimized Artificial Neural Network (ANN) modules for solar power, wind power, and electrical load demand forecasting, whose outputs are fused to derive the net load. Particle Swarm Optimization (PSO) is employed specifically to optimize the weights and biases of each ANN prior to the supervised learning phase, replacing conventional gradient-based initialization and thereby improving convergence characteristics and avoiding local minima entrapment. The parallel ensemble structure is motivated by the distinct statistical properties of each generation and consumption component, enabling specialized feature learning for each sub-problem before integration. Experimental validation using chronologically partitioned real-world operational data demonstrates that the proposed framework achieves MAPE values of 5.3%, 8.7%, and 3.8% for solar, wind, and load forecasting, respectively, and an overall net load MAPE of 6.2%, outperforming conventional methods including ARIMA, standard backpropagation ANN, and Support Vector Regression, as well as benchmark deep learning models. The results confirm that the PSO-optimized parallel ensemble approach provides a robust and computationally tractable solution for net load forecasting under high renewable penetration.