Developing an Integrated AI Model for Predictive Maintenance in Hydroponics

Hydroponics have gained attention due to their ability to cope with future food needs. However, farmers face the issue of maintenance: if done frequently, it is expensive, and if done rarely, it risks crop failure. This paper aims to ideate a setup of AI models for maintenance to be predictive, reducing crop failure risk while making maintenance more profitable. The model must predict future readings of various conditions that can indicate crop failure. Key factors identified include yield prediction, component malfunction, nutrition, electrical conductivity, power of hydrogen, and environmental factors. Results showed that deep neural networks are promising for yield prediction; random forest for component malfunction; random forest and support vector regression for nutrition; nonlinear autoregressive with exogenous inputs for EC and pH; and extreme gradient boosting for environmental factors. Deep neural networks can also be trained to mimic other models' decisions, lowering economic investment.