A Machine-Learning-Based IoT System for Optimizing Nutrient Supply in Commercial Aquaponic Operations

Abstract

Nutrient regulation in aquaponic environments has been a topic of research for many years. Most studies have focused on appropriate control of nutrients in an aquaponic set-up, but very little research has been conducted on commercial-scale applications. In our model, the input data were sourced on a weekly basis from three commercial aquaponic farms in Southeast Texas over the course of a year. Due to the limited number of data points, dimensionality reduction techniques such as pairwise correlation matrix were used to remove the highly correlated predictors. Feature selection techniques such as the XGBoost classifier and Recursive Feature Elimination with ExtraTreesClassifier were used to rank the features in order of their relative importance. Ammonium and calcium were found to be the top two nutrient predictors, and based on the months in which lettuce was cultivated, the median of these nutrient values from the historical dataset served as the optimal concentration to be maintained in the aquaponic solution to sustain healthy growth of tilapia fish and lettuce plants in a coupled set-up. To accomplish this, Vernier sensors were used to measure the nutrient values and actuator systems were built to dispense the appropriate nutrient into the ecosystem via a closed loop.

Keywords: ExtraTreesClassifier; Recursive Feature Elimination; XGBoost; aquaponic; closed loop; median; pairwise correlation matrix.

Figure 1

(a). set-up of the plant bed which was used for producing lettuce and other greens; and (b). a set-up of an array of interconnected fish tanks which served as a nutrient source for these plants in Texas US Farms, Grimes, TX, USA.

Figure 2

Feature selection pipeline for choice of appropriate nutrient predictors.

Figure 3

Pipeline of the complete system for sensing and actuation of nutrients for coupled systems in commercial aquaponic set-ups.

Figure 4

Pairwise correlation matrix among all the nutrient predictors in the dataset.

Figure 5

F-score of each predictor generated by the XGBoost algorithm.

Figure 6

Feature importance of each predictor generated by the ExtraTreesClassifier.

Figure 7

(a). A set-up of the Vernier sensors to measure calcium and ammonium; (b). a set-up of the motor control units and the MCUs to control the actuators; (c). a prototype of the full system showing the nutrients with the actuator units used to carry out their dispensation and (d). a feedback loop output communicating between the sensor and actuator modules.

Figure 8

Detailed block diagram of the IoT-based sensing and dispensing system.