An online machine learning-based sensors clustering system for efficient and cost-effective environmental monitoring in controlled environment agriculture

Abstract

Sensors are vital in controlled environment agriculture for measuring parameters for effective decision-making. Currently, most growers randomly install a limited number of sensors due to economic implications and data management issues. The microclimate within a protected cultivation system is continuously affected by the macroclimate (ambient), which further complicates decision-making around optimal sensor placement. The ambient weather's effect on the indoor microclimate makes it challenging to predict or acquire the ideal condition of the systems through using sensors. This study proposed and implemented a machine learning (K-Means++) algorithm to select optimal sensor locations through clustering. Temperature and relative humidity data were collected from 56 different locations within the greenhouse for over a year covering and these covered four major seasons (spring, summer, autumn, and winter). The data was processed to remove outliers or noise interference using interquartile. The original temperature and relative humidity data were transformed to other air properties (dew point temperature, enthalpy, humid ratio, and specific volume) and used in simulations. The results obtained showed that the number of optimal sensor locations ranged between 3 and 5, and there were similar sensor locations among the air properties. An online machine learning web-based system was developed to systematically determine the optimal number of sensors and location.