Assessing the stability of indoor farming systems using data outlier detection.

Abstract

Introduction: This study investigates the quality of air temperature data collected from a small-scale Controlled Environment Agriculture (CEA) system using low-cost IoT sensors during lettuce cultivation at four different temperatures. Ensuring data quality in CEA systems is essential, as it affects system stability and operational efficiency. This research aims to assess system stability by analyzing the correlation between cumulative agricultural operations (Agr.Ops) and air temperature data variability.

Methods: The methodology involved collecting air temperature data from IoT sensors in the CEA system throughout lettuce cultivation trials. A generalized linear model regression analysis was conducted to examine the relationship between cumulative Agr.Ops and the z-scores of air temperature residuals, which served as an indicator of system stability. Outliers in the sensor data were identified and analyzed to evaluate their impact on system performance. Residual distribution and curve fitting techniques were used to determine the best distribution model for the sensor data, with a log-normal distribution found to be the best fit.

Results: Regression analysis indicated a strong inverse relationship between cumulative Agr.Ops and residual z-scores, suggesting that increased Agr.Ops correlated with a higher presence of outliers and a decrease in system stability. The residual analysis highlighted that outliers could be attributed to potential issues such as sensor noise, drift, or other sources of uncertainty in data collection. Across different trials, the system displayed varying degrees of resistance to cumulative Agr.Ops, with some trials showing increased resilience over time.

Discussion: The alternative decomposition method used effectively identified outliers and provided valuable insights into the functionality of the system under different operational loads. This study highlights the importance of addressing uncertainties in indoor farming systems by improving surrogate data models, refining sensor selection, and ensuring data redundancy. The proposed method offers a promising approach for enhancing monitoring and managing uncertainties in CEA systems, contributing to improved stability and efficiency in indoor farming.