An LSTM-Based Online Prediction Method for Building Electric Load During COVID-19

Hao Tu, Srdjan Lukic, Abhishek Dubey, Gabor Karsai

Annual Conference of the PHM Society 2020

PDF

Why This Matters

Building energy consumption patterns change significantly in response to occupancy changes and operational modifications, making historical models ineffective when such changes occur. Traditional offline models cannot adapt to these concept changes, leading to poor predictions during transition periods. This work is innovative because it provides an online learning approach that enables energy prediction systems to adapt in real-time to concept changes, improving prediction accuracy and enabling better building management during periods of operational change.

What We Did

This paper presents an LSTM-based online prediction method for building electric load during COVID-19 that adapts to concept changes in energy consumption patterns. The approach uses online learning with adaptive learning rates to maintain model accuracy when building energy use patterns change fundamentally due to operational changes. The methodology includes ensemble approaches with multiple models at different learning rates to enable robust predictions despite changing consumption patterns.

Key Results

The LSTM-based approach successfully predicts building electric load and adapts to changes in energy consumption patterns caused by COVID-19 building closures and reopenings. Results show that online learning with adaptive learning rates improves prediction accuracy compared to fixed-rate online learning, achieving lower prediction errors as the system adapts to new patterns. The approach enables building management systems to maintain accurate energy predictions despite changing operational conditions.

Full Abstract

Cite This Paper

@inproceedings{haophm2020,
  author = {Tu, Hao and Lukic, Srdjan and Dubey, Abhishek and Karsai, Gabor},
  booktitle = {Annual Conference of the PHM Society},
  title = {An LSTM-Based Online Prediction Method for Building Electric Load During COVID-19},
  year = {2020},
  abstract = {Accurate prediction of electric load is critical to optimally controlling and operating buildings. It provides the opportunities to reduce building energy consumption and to implement advanced functionalities such as demand response in the context of smart grid. However, buildings are nonstationary and it is important to consider the underlying concept changes that will affect the load pattern. In this paper we present an online learning method for predicting building electric load during concept changes such as COVID-19. The proposed methods is based on online Long Short-Term Memory (LSTM) recurrent neural network. To speed up the learning process during concept changes and improve prediction accuracy, an ensemble of multiple models with different learning rates is used. The learning rates are updated in realtime to best adapt to the new concept while maintaining the learned information for the prediction.},
  contribution = {minor},
  tag = {ai4cps,power},
  keywords = {building energy prediction, LSTM, online learning, concept drift, adaptive learning, smart buildings}
}
Quick Info
Year 2020
Keywords
building energy prediction LSTM online learning concept drift adaptive learning smart buildings
Research Areas
energy ML for CPS
Search Tags

LSTM, Online, Prediction, Method, Building, Electric, Load, During, COVID, building energy prediction, online learning, concept drift, adaptive learning, smart buildings, energy, ML for CPS, 2020, Tu, Lukic, Dubey, Karsai