(207h) Simulated Greenhouse Lighting Control Test Bed Utilizing Ray Tracing and Image Detection to Study Control and Modeling Considerations Using Classical and Quantum Computing

Energy-efficient greenhouses are desirable to reduce carbon dioxide emissions, cut expenditures, and increase crop yields. Lighting costs constitute a significant portion of the budgets of greenhouses [1, 2], motivating the application of new lighting control method to limit energy consumption. Many control architectures have been proposed. This includes, for example, methods of manipulating LEDs to track a desired energy flux set points [3, 4] or though modeling and tracking aggregate properties such as the dry weight per area [5], utilizing a variety of control algorithms such as model predictive control or machine learning [3]. Modern sensing methods such as image-based sensing could also be desirable to incorporate. One of the challenges for a control engineering research in attempting to evaluate advanced control for applications such as these could be the lack of a greenhouse to experiment with.

Recent work in our group has been exploring the potential of the computer graphics software Blender for applications involving image-based control (e.g., [6]). Blender includes a 3D modeling interface, as well as capabilities for rendering and for coding in a Python interface. It therefore is a potentially interesting test bed for a variety of applications, including cases such as evaluating advanced control of systems such as greenhouses, where physics-only simulations such as are common in process systems engineering (e.g., modeling only temperatures, pressures, or concentrations which are not required to be visualized and where the physics are well-characterized) may not be suitable. It could allow, for example, testing of an algorithm for controlling greenhouse lighting for a variety of different plant growth models, to enable understanding of how the algorithms might behave for a variety of different scenarios even if the plant growth model is not fully known.

For these reasons, in this work we create a test bed for control of a greenhouse lighting system. The test bed is implemented in Blender using the Python programming language. Blender is used to generate the structure of leaves in the 3D modeling interface. The leaves represent plants in the ‘actual’ greenhouse. The positions of the leaves are part of the feedback received by the controller that manipulates the LEDs to promote the plant growth. The leaf positions are read using a method of processing a point cloud representation of the object (as in [7]) obtained from the image of the plant rendered using Blender into a 3D reconstruction of the object. Next, a model predictive controller equipped with a simplified ray tracing model is used to selectively manipulate individual LEDs to minimize energy costs while ensuring each leaf maintains a desired energy flux. Additionally, the models consider sunlight and plant growth which act as disturbances. Closed-loop simulation results are presented to demonstrate the performance of the model and optimization-based controller, and we discuss methods that can be applied to improve controller performance.

Next, we demonstrate how the test bed can be used to analyze the application of hybrid classical-quantum algorithms that could be of interest for future applications of the greenhouse test bed to better understand how quantum computation [8] might interact with it. To do this, we describe the potential use of algorithms such as the quantum random walk for applications in representing stochastic plant growth in greenhouses. The quantum random walk has the potential to find the optimal path in a graph of vertices and edges, where the edges represent a probability of moving from one vertex to the next [9]. Stochastic growth models represent the growth of the geometry of a plant based on a set of probabilistic rules [10, 11]. We represent a stochastic growth model as a graph of vertices and edges, and then determine the optimal path using a quantum random walk algorithm, for use in determining the direction of plant growth. The results of this analysis are then compared to the implementation of a purely classical test bed. Through these simulations, we demonstrate the capabilities of Blender for a variety of non-traditional advanced control applications where test beds may otherwise be difficult to find without an experimental greenhouse system.

[1]. van Iersel, Marc W., and David Gianino. "An adaptive control approach for light-emitting diode lights can reduce the energy costs of supplemental lighting in greenhouses." HortScience 52.1 (2017): 72-77.

[2]. Katzin, David, Leo FM Marcelis, and Simon van Mourik. "Energy savings in greenhouses by transition from high-pressure sodium to LED lighting." Applied Energy 281 (2021): 116019.

[3]. Mohagheghi, Afagh, and Mehrdad Moallem. "An Energy-Efficient PAR-Based Horticultural Lighting System for Greenhouse Cultivation of Lettuce." IEEE Access (2023).

[4]. Jiang, Jun, Afagh Mohagheghi, and Mehrdad Moallem. "Energy-efficient supplemental LED lighting control for a proof-of-concept greenhouse system." IEEE Transactions on industrial electronics 67.4 (2019): 3033-3042.

[5]. van Mourik, Simon, Bert van'T. Ooster, and Michel Vellekoop. "Plant Performance in Precision Horticulture: Optimal climate control under stochastic uncertainty." arXiv preprint arXiv:2303.14678 (2023).

[6] Oyama, Henrique, Dominic Messina, Keshav Kasturi Rangan, Akkarakaran Francis Leonard, Kip Nieman, Helen Durand, Katie Tyrrell, Katrina Hinzman, and Michael Williamson. "Development of directed randomization for discussing a minimal security architecture." Digital Chemical Engineering 6 (2023): 100065.

[7]. Berger, Matthew, et al. "A survey of surface reconstruction from point clouds." Computer graphics forum. Vol. 36. No. 1. 2017.

[8]. Yanofsky, Noson S., and Mirco A. Mannucci. Quantum computing for computer scientists. Cambridge University Press, 2008.

[9]. Koch, Daniel, and Mark Hillery. "Finding paths in tree graphs with a quantum walk." Physical Review A 97.1 (2018): 012308.

[10]. Kang, Meng-Zhen, et al. "Analytical study of a stochastic plant growth model: Application to the GreenLab model." Mathematics and Computers in Simulation 78.1 (2008): 57-75.

[11]. Kang, Mengzhen, et al. "Parameter identification of plant growth models with stochastic development." 2016 IEEE International Conference on Functional-Structural Plant Growth Modeling, Simulation, Visualization and Applications (FSPMA). IEEE, 2016.