Conference Agenda

World population is expected to reach nearly 10 billion people by 2050. Inequalities in food access will continue to be exacerbated without the intervention of advanced solutions. Vertical farming is a technology field which could allow for the expansion of food production by stacking garden spaces on top of one another. From an industrial perspective, this is a practical method for optimizing square footage of farms without interrupting native plant ecosystems or suburban and business properties. From a small-scale perspective, vertical farms may also be implemented into homes enabling homeowners to become prosumers as they help to decentralize the global food chain. At Purdue University’s DC Nanogrid House, a Vertical Farm (VF) has been constructed to prove the feasibility and effectiveness of the agricultural technology.

While other papers focus on the global scale of VF and how it will determine future commercial and research farming, the goal of the present study is to develop an automated VF system for a residential application. Autonomous irrigation is installed and controlled via ESP-32 Micro Controller and capacitive soil moisture sensors, and high-efficiency LED lights are controlled based on the plants’ needs. Methods for efficient power use, reduced water use, fertilizing, and other relevant topics will be discussed for the user’s benefit. Readers should come away from this paper with an understanding of the benefits of VF and a plan for creating an at-home VF with an automatic irrigation system.

Accurate prediction of solar irradiance components plays a pivotal role in optimizing buildings and solarsystems performance, in particular in areas characterized by shading and terrain effects such as mountain regions.

This is generally conducted through statistical regression of collected data series. When faced with a dataset consisting of inputs, Xi, in n-dimensional space and their corresponding responses, Yi, in real numbers, symbolic regression aims to find a function, f, that accurately fits the dataset while providing a concise and comprehensible mathematical expression (Alaoui Abdellaoui and Mehrkanoon, 2021).

The mathematical expressions in this context exist within a unique space characterized by a discrete functional form but a continuous parameter space, encompassing floating-point constants and variables. As the length of these expressions grows, the complexity of the symbolic regression's search space expands exponentially. This presents a significant challenge for symbolic regression within the field of machine learning (Udrescu and Tegmark, 2020).

Due to the vast and combinatorial nature of the search space, traditional approaches to symbolic regression often turn to evolutionary algorithms (Eiben and Smith, 2003), with the genetic algorithm (GA) (Banzhaf et al., 1998) emerging as a prominent choice. In GA-based symbolic regression, a population of mathematical expressions evolves through iterations, with selection favoring expressions that fit the data based on a fitness function. Selected expressions undergo crossover to create diverse offspring, and some offspring experience random mutation. Guided by the fitness function, this iterative process refines expressions to find the optimal, interpretable expression that fits the data. This methodology provides an effective means of navigating the intracity of mathematical expressions to unveil optimal ones (Ruggiero, 2020).

Accordingly, the study aims at generating mathematical expressions through a two-step process. Initially, the Random Forest (RF) regressor (Breiman, 2001) allows dividing the dataset into tree-based subsets. A GA-based symbolic regression methods is then applied, as presented by Stephens (2016) and Banzhaf et al. (1998), to uncover mathematical expressions within these subsets. This simplifies the complexity of the search space in the GA-based symbolic regression model, reducing computational demands and establishing precise relationships between predictors and targets within each subset.

The methodology allows predicting global horizontal irradiance (GHI) and global tilted irradiance (GTI) for various cardinal orientations at hourly intervals in complex environments, including urban and mountainous regions generating precise mathematical expressions that capture complex relationships, even in shaded areas.

Alaoui Abdellaoui, I., & Mehrkanoon, S. (2021). Symbolic regression for scientific discovery: an application to wind speed forecasting. arXiv preprint arXiv:2102.10570.

Banzhaf, Wolfgang; Nordin, Peter; Keller, Robert; Francone, Frank (1998). Genetic Programming – An Introduction. San Francisco, CA: Morgan Kaufmann. ISBN 978-1558605107.

Breiman, L. (2001). Random Forests. Machine Learning, 45, 5–32.

Eiben, A.E., Smith, J.E. (2003). Introduction to Evolutionary Computing. Springer, Berlin, Heidelberg. DOI: https://doi.org/10.1007/978-3-662-44874-8-

Ruggiero, R. (2020). Symbolic Regression: The Forgotten Machine Learning Method. Towards Data Science. Retrieved from https://towardsdatascience.com/symbolic-regression-the-forgotten-machine-learning-method-ac50365a7d95

Stephens, T. (2016). Gplearn: Genetic Programming for Symbolic Regression [GitHub repository].

Udrescu, S. M., & Tegmark, M. (2020). AI Feynman: a Physics-Inspired Method for Symbolic Regression. Science Advances, 6(16), eaay2631

Heat pump water heaters (HPWHs), stand alone, are much more efficient than gas or electric resistance water heaters but have poor performance at low ambient air temperatures. Furthermore, when installed in small mechanical closets commonly found in low-income multifamily residences, unitary air-source HPWHs take usable heat from space and result in higher space heating loads in winter, especially in cold climates where the water heating loads tend to be higher than in warmer climates. In multifamily buildings served by centralized HPWHs, air-source (AS) HPWHs require make-up air that comes from the outdoors that can significantly impact the capacity and performance of the HPWH as well. Therefore, the energy burden for low-income families in cold climates may not be solved by incentivizing air-source water heaters when improperly installed in small closets or requiring significant amounts of outdoor air. A new source of heat at higher temperatures would significantly improve the COP of HPWHs in cold climates and drain wase heat recovery is the solution (DWHR).

In this work, a novel system has been proposed to overcome the aforementioned challenges in multifamily buildings in cold climates. The proposed Hybrid Centralized & Distributed water heating improves the centralized systems by including a water-source (WS) HPWH typically used for geothermal applications. The system includes the “regenerative braking of water heating” DWHR with a water-source HPWH as the source of heat on the evaporator side of the HPWH. The DWHR harvests energy from the drain while the temperatures are warm enough (i.e. > 60 °F) and stores it in a large moderate-temperature storage tank in which the heat is then available for lifting to domestic hot water temperatures by the WS-HPWH. Since the source temperature for the WS-HPWH is >> than the AS-HPWH high performance and capacity is realized. Furth more, small tanks a distributed to each apartment to reduce the amount of heat lost in the centralized hot-water recirculation loop.

A technoeconomic study was conducted for centralized air-source HPWH and drain-source HPWH technologies, in the traditionally centralized and C&D configuration, based on model performances in cold climates. The results are then compared to traditional water heating solutions with projections into the future based on 5 difference scenarios. The analysis shows a yearly coefficient of performance (COP) greater than 4 is achieved, the technology is cost competitive against all other water heating technologies.

This work presents a fast and simplified model designed to simulate the temporal evolution of temperature and humidity within an industrial plant in Spain. The primary objective is to provide a comprehensive understanding of the thermal dynamics within the facility to improve the comfort of the staff, especially in summer as the reached temperatures are quite high. To achieve this, the model incorporates external temperature and solar radiation historical data as boundary conditions, effectively capturing the influence of external factors on the plant's climate over time.

The simulation employs transient box models to solve the heat equation, offering a practical approach to predicting temperature and humidity variations within the industrial setting. A 1D transient equation is used to address the complexities of different walls and layers of materials, considering factors such as orientation and solar radiation impact. Additionally, the model accounts for the inclusion of glass surfaces, that have a great impact in the studied plant.

The ventilation system is a critical aspect of the simulation, modeled with a similar control methodology as the overall plant. The plant does not have air conditioning and uses a free cooling approach for the summer. Furthermore, the model includes the internal dynamics of the industrial plant, incorporating the heat and humidity generation from internal machines, and the thermal inertia of possible internal elements.

In summary, this work provides a valuable contribution by offering a fast and simplified simulation model that captures the temporal evolution of temperature and humidity in an industrial plant. The inclusion of external conditions, detailed treatment of walls and windows, modeling of the ventilation system, and consideration of internal machine dynamics collectively contribute to a more accurate representation of the complex thermal interactions within the industrial environment.

Future weather data plays a critical role in assessing the impacts of climate change on building performance. However, the building energy modeling community has been struggling with a lack of high-quality future weather datasets tailored for building energy modeling, which leads to widespread misprocessing and misuse of future weather data in the field. To address this issue, we present a concise overview of the methodology used to create future weather data, covering aspects such as emissions scenarios, general circulation models, downscaling techniques, and the various types of future weather data. Additionally, we introduce a dynamically downscaled future weather dataset developed by Argonne National Laboratory, highlighting its superior qualities through a case study.