Scholarlink

<p>The escalating global demand for electricity and the growing need for sustainable energy practices underscore the importance of accurate load forecasting, particularly within energy-intensive environments like educational institutions. This study presents a robust deep learning framework for predicting laboratory-level energy consumption using smart meter data from four electrical labs at Thiagarajar College of Engineering, India. By integrating Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and a hybrid LSTM-GRU model, the study leverages temporal dependencies to forecast load profiles with high precision. The models are statistically validated through Augmented Dickey-Fuller tests and autocorrelation analysis, ensuring data stationarity and randomness readiness. Exploratory Data Analysis (EDA) is employed to uncover energy usage patterns, peak load periods, and seasonal variations. A comparative evaluation of various normalization techniques in conjunction with the Adam optimizer reveals that Z-Score normalization achieves the best performance, with a maximum <i>R</i><sup>2</sup> score of 0.8645 and minimum RMSE of 0.03485. The hybrid model outperforms standalone approaches, demonstrating its suitability for capturing complex consumption trends. This research contributes a scalable, data-driven methodology for load forecasting, enabling improved demand-side management, enhanced energy efficiency, and informed integration of renewable resources within higher educational infrastructures.</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

Building LSTM based forecast model with pattern exploration of load profiles for enhancing energy efficiency in electrical laboratories

  • 期刊论文

  • 发表日期：

    2025年6月10日

查看全文
  • J. Roshan Joshua,
  • S. G. Shiva,
  • S. Charles Raja,
  • A. C. Vishnu Dharssini

摘要

The escalating global demand for electricity and the growing need for sustainable energy practices underscore the importance of accurate load forecasting, particularly within energy-intensive environments like educational institutions. This study presents a robust deep learning framework for predicting laboratory-level energy consumption using smart meter data from four electrical labs at Thiagarajar College of Engineering, India. By integrating Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and a hybrid LSTM-GRU model, the study leverages temporal dependencies to forecast load profiles with high precision. The models are statistically validated through Augmented Dickey-Fuller tests and autocorrelation analysis, ensuring data stationarity and randomness readiness. Exploratory Data Analysis (EDA) is employed to uncover energy usage patterns, peak load periods, and seasonal variations. A comparative evaluation of various normalization techniques in conjunction with the Adam optimizer reveals that Z-Score normalization achieves the best performance, with a maximum R2 score of 0.8645 and minimum RMSE of 0.03485. The hybrid model outperforms standalone approaches, demonstrating its suitability for capturing complex consumption trends. This research contributes a scalable, data-driven methodology for load forecasting, enabling improved demand-side management, enhanced energy efficiency, and informed integration of renewable resources within higher educational infrastructures.