In recent decades, there has been a gradual penetration of plant factories achieving semiautomated crop cultivation. However, efficient energy utilization, as well as quality control of crops, are very important factors with regard to sustainable operation. Operating parameters, such as room temperature, affect not only the quality of crops but also the electric power required to realize the target operation while being influenced by the environment outside the plant. Therefore, a methodology is needed to analyze and interpret the relationships among these manipulated variables, exogenous variables, crop quality, and the amount of required electric power. Constructing a directed acyclic graph composed of regression models is an attractive approach for such analysis; however, the relationships can possibly be nonlinear, so the direct application of existing analytic approaches will not be appropriate. In this paper, we propose a methodology for relationship analysis among variables based on the directed acyclic graphs while identifying the linearity/nonlinearity in their relationships. In general, the construction of such a graphical model has computational issues, especially when the number of variables is large, and the risk of overfitting. The proposed method utilizes the idea of sparse regularization, which has been actively discussed in the field of machine learning, for realizing the automatic identification of linearity/nonlinearity between variables and screening redundant candidate structures; this approach relaxes the computational complexity issue and controls the risk of overfitting. As a case study, the proposed method is applied to a dataset collected from a real-world cultivation system in a plant factory to discuss its usefulness.
ASJC Scopus subject areas