A Seasonal-Trend Decomposition and Single Dendrite Neuron-Based Predicting Model for Greenhouse Time Series

The greenhouse farming always uses sensors to monitor the dynamic climate parameters and generate time-related data. The prediction of these time series contributes a lot to greenhouse cultivation. Plenty of works concentrate on the chaotic characteristics of the time series and propose many machine learning-based models. However, the intrinsic features of them are ignored, i.e., seasonality and tendency. In this study, we propose a novel predicting model SDN that utilizes the Seasonal-trend Decomposition as preprocessing method and the Single Dendrite Neuron as nonlinear fitter to tackle greenhouse time series predictions. The decomposition gives SDN a flexibility that can process each component separately, while the well-designed neuron structure provides SDN with time efficiency. Accordingly, the experimental results show that the proposed SDN not only beats the widely used machine learning-based models, but also shows the robustness considering customized parameters and outliers in datasets, which enhance the possibility for SDN to be employed in the practical usage scenarios.