Project 4: Short-Term Load Forecasting Using Machine Learning

This project explores short-term load forecasting (STLF) using deep learning models to predict electricity demand in buildings, leveraging the BuildingsBench platform developed by the National Renewable Energy Laboratory. 

Project Summary

Short-Term Load Forecasting (STLF) is the process of predicting the electrical energy demand of residential and commercial buildings over short timeframes, ranging from the next hour to the next month. STLF plays a vital role in efficient energy management, making it an important research area in the United States government's goal of achieving energy dominance. We leverage the BuildingsBench platform, developed by the National Renewable Energy Laboratory, a U.S. Department of Energy lab, to explore the role of deep learning in STLF. 

In this project, we will build and train recurrent neural networks (RNNs), which are well-suited for handling time-series data and predicting long-term trends, to forecast electricity load for a variety of buildings. This project will provide you with hands-on experience with object-oriented programming in Python, as well as working with the PyTorch, mpi4py, and NumPy libraries, tuning model parameters, and training RNNs on multiple GPUs. This project emphasizes practical implementation with minimal focus on theoretical foundations. Prior coding experience in Python is a plus, but no additional background is required.

Learning Objectives

  1. Learn to visualize time-series data using Python.
  2. Design RNNs using PyTorch
  3. Train and tune the RNNs on multiple GPUs
  4. Evaluate the trained model’s accuracy
  5. Analyze model outcomes

Project Leads

Peer Mentors

Materials

All Jupyter notebooks, data files, and other required materials for Project 4 can be found at the Github repository link below.

Project 4 Github repo