Tech Insights

Revolutionizing Energy Demand Estimation With Machine Learning

April 14, 2023 by Claire Turvill

Researchers from University College Cork and Columbia University have published a study using TrebuNet, a machine learning technique, to better project future demand in the transportation industry.

According to the United Nations, the population worldwide is expected to reach approximately 9.7 billion people by 2050–a 21 percent increase. This growth is expected to escalate the demand for services across various industries. Consequently, University College Cork (UCC) and Columbia University have published research to enhance the capacity to project future demand in the transportation sector.

 

New research aims to project future demand in the transportation industry better. Image used courtesy of Pixabay

 

Environmental Impacts

The researchers have implemented machine learning techniques to improve the accuracy of estimating the future demands for passenger and freight transportation, which collectively contribute to 20 percent of the world’s greenhouse gas emissions.

The transportation industry is the highest greenhouse gas emitter because it relies heavily on fossil fuels, mainly oil-based products such as gasoline and diesel fuel, to power vehicles. The demand for transportation services continues to increase globally due to population growth, urbanization, and globalization, which put additional pressure on the transportation sector to supply these services.

Numerous initiatives are being implemented to minimize the carbon footprint of the transportation industry by adopting cleaner and more sustainable modes of transportation, including electric vehicles, public transportation, and alternative fuels.

Accurately analyzing the energy demands of the transportation sector is crucial for mitigating climate change while also addressing one of the fundamental needs of humanity–mobility.

The research conducted by UCC and Columbia endeavors to leverage machine learning to refine transport demand projections, thereby providing improved support for energy system models and climate policies alongside other measures to reduce emissions.

 

Predicting Demand With TrebuNet

Previously, estimating transportation demand projections involved simulations or regression-based analysis. However, researchers from UCC and Columbia have found a way for countries worldwide to estimate future transport demands more accurately.

In this study published in Scientific Reports, a novel deep learning neural network architecture named TrebuNet is introduced to improve the accuracy of estimating transport energy service demand. TrebuNet is designed to mimic the physical process of firing a trebuchet, thereby modeling the nuanced dynamics inherent in energy service demand estimation. 

The study outlines the design, training, and implementation of TrebuNet and compares its performance to traditional multivariate linear regression and state-of-the-art machine learning algorithms such as densely connected neural networks, recurrent neural networks, and gradient-boosted methods. The results demonstrate that TrebuNet outperforms these methods in regional demand projection for all modes of transport demands at short, decadal, and medium-term time horizons. 

 

Learning Phase and Firing Phase of the TrebuNet Machine Learning. Image used courtesy of Scientific Reports

 

The TrebuNet concept includes a physical learning process during the launch of a projectile and firing it accurately after learning. The Learning Phase of TrebuNet generates different projections for different quantiles, while the Firing Phase combines different quantiles into one accurate projection based on errors in different quantile projections.

Additionally, the study presents a framework for projecting energy service demand in regions with multiple countries spanning different socio-economic development pathways, which can be replicated for wider regression-based tasks for time series with non-uniform variance.

This research offers critical insights into creating a new machine-learning architecture that enhances the accuracy of estimating transport energy service demands. The benefits of this innovative machine learning architecture are palpable for the energy modeling community and have broad applicability across diverse disciplines.

Precise transport demand projections are crucial not only for energy system models and climate policies but also as the fundamental basis for comprehending the future course of global energy markets.

The introduction of this novel approach showcases innovation in energy systems modeling and data analytics by addressing the limitations in comprehending the prospects of new deep learning applications within energy system models. This enables us to eliminate uncertainties in decarbonization pathways.

Achieving the target of global net-zero emissions by 2050 demands prompt climate action to decarbonize the transportation sector. The partnership between Columbia and UCC is spearheading pioneering techniques in energy systems modeling and data science to equip decision-makers with evidence-based research and tools to design effective climate policies.