In late 2019, ASHRAE hosted the Great Energy Predictor III (GEPIII) machine learning competition on the Kaggle platform. This launch marked the third energy prediction competition from ASHRAE and the first since the mid-1990s. In this updated version, the competitors were provided with over 20 million points of training data from 2,380 energy meters collected for 1,448 buildings from 16 sources. This competition’s overall objective was to find the most accurate modeling solutions for the prediction of over 41 million private and public test data points. The competition had 4,370 participants, split across 3,614 teams from 94 countries who submitted 39,403 predictions. In addition to the top five winning workflows, the competitors publicly shared 415 reproducible online machine learning workflow examples (notebooks), including over 40 additional, full solutions. This paper gives a high-level overview of the competition preparation and dataset, competitors and their discussions, machine learning workflows and models generated, winners and their submissions, discussion of lessons learned, and competition outputs and next steps. The most popular and accurate machine learning workflows used large ensembles of mostly gradient boosting tree models, such as LightGBM. Similar to the first predictor competition, preprocessing of the data sets emerged as a key differentiator.
One of the critical biotic stress factors paddy farmers face is diseases caused by bacteria, fungi, and other organisms. These diseases affect plants’ health severely and lead to significant crop loss. Most of these diseases can be identified by regularly observing the leaves and stems under expert supervision. In a country with vast agricultural regions and limited crop protection experts, manual identification of paddy diseases is challenging. Thus, to add a solution to this problem, it is necessary to automate the disease identification process and provide easily accessible decision support tools to enable effective crop protection measures. However, the lack of availability of public datasets with detailed disease information limits the practical implementation of accurate disease detection systems. This paper presents Paddy Doctor, a visual image dataset for identifying paddy diseases. Our dataset contains 16,225 annotated paddy leaf images across 13 classes (12 diseases and normal leaf). We benchmarked the Paddy Doctor dataset using a Convolutional Neural Network (CNN) and four transfer learning based models (VGG16, MobileNet, Xception, and ResNet34). The experimental results showed that ResNet34 achieved the highest F1-score of 97.50%. We release our dataset and reproducible code in the open source for community use.
Maintaining both indoor air quality (IAQ) and thermal comfort in buildings along with optimized energy consumption is a challenging problem. This investigation presents a novel design for hybrid ventilation system enabled by predictive control and soft-sensors to achieve both IAQ and thermal comfort by combining predictive control with demand controlled ventilation (DCV). First, we show that the problem of maintaining IAQ, thermal comfort and optimal energy is a multi-objective optimization problem with competing objectives, and a predictive control approach is required to smartly control the system. This leads to many implementation challenges which are addressed by designing a hybrid ventilation scheme supported by predictive control and soft-sensors. The main idea of the hybrid ventilation system is to achieve thermal comfort by varying the ON/OFF times of the air conditioners to maintain the temperature within user-defined bands using a predictive control and IAQ is maintained using Healthbox 3.0, a DCV device. Furthermore, this study also designs soft-sensors by combining the Internet of Things (IoT)-based sensors with deep-learning tools. The hardware realization of the control and IoT prototype is also discussed. The proposed novel hybrid ventilation system and the soft-sensors are demonstrated in a real research laboratory, i.e., Center for Research in Automatic Control Engineering (C-RACE) located at Kalasalingam University, India. Our results show the perceived benefits of hybrid ventilation, predictive control, and soft-sensors.
This study addresses the urgent need for comprehensive climate education amid a climate emergency. Human (energy) behaviors are developed from childhood and early adulthood. This study hypothesizes that transcending a nation’s net-zero energy ambition can be accomplished through experiential education. An Urban Governance Lab plus nEt-Zero Energy league model is introduced. Various behavioral interventions are designed based on the principles of serious games. Discussions provide rich narratives on how a nation with so many diverse communities can forge a rapid net-zero transition. The blended multi-disciplinary STEM education can drive energy citizenship in campus-like communities. A scenarios-based analysis demonstrating the potential of the proposed model in shaping energy behavior in young citizens leading to net zero is presented. The results from the scenario analysis present optimistic evidence underlining how campus-like communities driven by bottom-up initiatives can realize net-zero ambition beyond hope.
The recent advancements in Time Series Foundation Models (TSFMs), which are pre-trained on vast time series corpus from diverse domains and designed to handle universal forecasting tasks without the need for task-specific training, present a promising domain-agnostic, one-size-fits-all solution. In this paper, we analyze the performance of four open-source TSFMs - Chronos, Lag-Llama, Moirai, and TimesFM - for short-term load forecasting (STLF) in both commercial and residential buildings. Specifically, we benchmark these models' zero-shot prediction capabilities, assessing their ability to perform STLF on unseen buildings, and compare them to state-of-the-art models. Our experimental results, conducted on a large-scale dataset of over 1,900 real-world residential and commercial buildings, reveal that the Chronos model achieves the lowest Normalized Root Mean Squared Error (NRMSE) compared to the contemporary models and demonstrates comparable performance with an existing pre-trained TSFM for the energy domain from the BuildingsBench platform.
Accurate short-term load forecasting (STLF) is essential for energy management in buildings and promoting sustainability. While various methods---ranging from statistical to machine learning---have been proposed, their evaluation often focuses on a limited number of buildings, hindering scalability and generalizability. Recent advancements in Time Series Foundation Models (TSFMs), pre-trained on extensive time series data from diverse domains, offer a promising domain-agnostic solution that can handle universal forecasting tasks without requiring task-specific training. In this paper, we analyze the performance of four open-source TSFMs - Chronos, Lag-Llama, Moirai, and TimesFM - for STLF in both commercial and residential buildings. Specifically, we benchmark these models' zero-shot prediction capabilities, assessing their ability to perform STLF on unseen buildings, and compare them to state-of-the-art models. Our experimental results, conducted on a large-scale dataset of over 1,900 real-world residential and commercial buildings, reveal that the Chronos model achieves the lowest Normalized Root Mean Squared Error compared to the contemporary models and demonstrates comparable performance with an existing pre-trained TSFM for the energy domain from the BuildingsBench platform. Moreover, our analysis of the forecasting errors of Chronos reveals no significant deviation in errors across different building types. Finally, we share insights to further improve the performance of TSFMs for STLF tasks.