Our Lab's New AI Model Tackles Missing Data in Building Energy Systems
We are excited to announce the publication of our latest research, ‘Adaptive neural temporal hybridization for missing data imputation in building energy use datasets: An integrated LNN- LSTM weighted model’, in the prestigious Journal of Building Engineering. This work, led by Saeed Murtaza, Sarath Raj, and Prof. Geun Young Yun, introduces a groundbreaking approach to a persistent challenge in building energy management.
The Problem: Incomplete Data
Effective energy management in buildings relies on accurate, continuous data. However, due to sensor malfunctions or communication errors, energy consumption data is often incomplete. These gaps can significantly impair the performance of data-driven systems used for forecasting energy demand and optimizing efficiency.
Our Solution: The Integrated LNN-LSTM Weighted Model (ILLWM)
To address this challenge, our research team developed a novel deep learning model called the Integrated LNN-LSTM Weighted Model (ILLWM). This innovative model combines the strengths of two powerful neural networks:
● Liquid Neural Networks (LNN): Known for their flexibility and ability to adapt to dynamically changing data streams.
● Long Short-Term Memory (LSTM) Networks: Excel at recognizing and modeling long- term patterns in time-series data.
The ILLWM uses an intelligent, RMSE-based weighting system to dynamically combine the outputs of both LNN and LSTM, leveraging the best capabilities of each to produce a single, highly accurate estimate for the missing data points. Key Findings and Impact
We tested the ILLWM on real-world energy consumption data from commercial, residential, and hospital buildings, even with high rates of missing data (up to 40%). The results were remarkable.
Our model significantly outperformed existing imputation methods. For instance, in a commercial building dataset with 40% missing data, the ILLWM reduced imputation errors by 89.6% compared to a standard LSTM model and 76.9% compared to a standalone LNN model.
By ensuring more complete and reliable datasets, the ILLWM can directly enhance the accuracy of energy demand forecasting and support the development of more sustainable and efficient energy management strategies in buildings worldwide.
Read the Full Paper
We congratulate the authors on this significant contribution to the field.
Authors: Saeed Murtaza, Sarath Raj, Geun Young Yun, Duk-Joon Park, Ji-Hye Kim, Gwanyong Park, and Jin Woo Moon.
The full paper is available in the Journal of Building Engineering.
