A WSN-based prediction model of microclimate in a greenhouse using an extreme learning approach

Monitoring and controlling microclimate in a greenhouse becomes one of the research hotspots in the field of agrometeorology, where the application of Wireless Sensor Networks (WSN) recently attracts more attentions due to its features of self-adaption, resilience and cost-effectiveness. Present microclimate monitoring and control systems achieve their prediction by manipulating captured environmental factors and traditional neural network algorithms; however, these systems suffer the challenges of quick prediction (e.g. hourly and even minutely) when a WSN network is deployed. In this paper, a novel prediction method based on an Extreme Learning Machine (ELM) algorithm is proposed to predict the temperature and humidity in a practical greenhouse environment in Nanjing, China. Indoor temperature and humidity are measured as data samples via WSN nodes. According to the results, our approach (0.0222s) has shown significant improvement on the training speed than Back Propagation (BP) (0.7469s), Elman (11.3307s) and Support Vector Machine (SVM) (19.2232s) models, plus the accuracy rate of our model is higher than those models. In the future, research on faster learning speed of the ELM based neural network model will be conducted.

I. Introduction

Modern greenhouses provide a suitable indoor microclimate meeting the requirements of plant growth. A prediction model of the microclimate in a greenhouse therefore becomes critical for the establishment of control strategies and consequent evaluation [1]. The design of such a model becomes challenging due to the features of the greenhouse microclimate, i.e. nonlinear, multiple input multiple output, and its strong coupling between relevant factors. It is also affected by the indoor and outdoor climate environment, crops grown inside and movements of control facilities [2]. All reasons above make it difficult to establish a precise mathematical model to achieve fast and accurate prediction [3]. With rapid development of short-range wireless communication, e.g. Wireless sensor networks (WSN), realtime (or nearly real-time) collection of relevant environmental data in a greenhouse turns out to be convenient, but also raises new challenges on microclimate prediction [4].