Infrastructure Systems: Machine Learning Techniques for Energy Forecasting and Optimization

Led by Zhen Ni, Ph.D.

REU Scholar: Mahim Rahaman

REU Scholar Home Institution: Lehman College

REU Mentor: Zhen Ni, Ph.D.

Enhancing Energy Management: Advanced Techniques for Forecasting and Optimization

This project focused on data visualization, computer programming, and result analysis using artificial intelligence and machine learning for energy forecasting and management. Leveraging smart energy data, including electricity demand and load profiles, we aimed to understand and predict energy consumption patterns. Extensive data visualization was conducted to uncover trends, followed by thorough data cleaning. Machine learning algorithms, specifically Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN), were utilized to forecast energy consumption. To further enhance our analysis, we employed the Transformer model, an advanced algorithm that demonstrated superior performance in handling time series data compared to traditional methods. The importance of this project lies in its potential to significantly improve energy management systems. Accurate energy consumption predictions enable more efficient energy distribution, reducing costs and environmental impact. By demonstrating the advantages of advanced algorithms like Transformers, this research highlights the potential for technological advancements in optimizing energy use, which is crucial in the context of growing energy demands and the need for sustainable practices

REU Scholar: Matthew Orellana

REU Scholar Home Institution: Florida Atlantic University

REU Mentor: Zhen Ni, Ph.D.

Load Demand Forecasting: A Comparative Study of Deep Learning Models

Electrical energy consumption is a crucial aspect of daily life, and accurate prediction is essential. Misestimating consumption can lead to overproduction, causing blackouts, or underproduction, resulting in unnecessary costs for households and businesses. Accurate predictions help prevent these issues by ensuring a sufficient and cost-effective energy supply. This project is focused on creating deep learning models to predict energy consumption during the summer of 2017 in Tetouan, Morocco. These models were tested for accuracy by applying them to previous years' seasonal data. Various models were explored, including recurrent neural networks (RNNs), deep neural networks (DNNs), SHAP (Shapley Additive Explanations), and Fourier series transformation. The findings indicate that the DNN model performed particularly well, achieving a mean absolute percentage error (MAPE) of 1% for the Tetouan summer data. While the model also showed promise with other datasets, it was most accurate for Tetouan. These results suggest that with further training and optimization, the model could predict energy consumption with even greater accuracy.