A Multi-Layer Perceptron Neural Network Approach for Greenhouse Protection against Strong Wind Conditions in the Southern Coastal Regions of China

Abstract

In the coast regions of China, strong wind activities such as typhoon are frequent.Typhoon is a major threat to greenhouses. During the typhoon periods, energy level of the winds isusually erratic, sometimes weak and sometimes strong. The nave approach of controllinggreenhouse wind protection shutters according to the current wind speed will fail due to multivariatefactors that affect wind speed. In this paper, we study the dynamic properties that cause the nonlinearperiodic change of wind speed. A Multi-Layer Perceptron Neural Network model is designed toautomatically acquire a classification model for real-time strong wind detection, using raw sensoryreading of wind speed as training data. This work provides an approach for the unification ofagricultural production and management control. Results show that this system is sufficient for thestrong wind protection requirement of the south china coastal regions. Experiments also demonstratethat the system is robust against erratic strong wind behaviors that are frequent under real typhoonconditions.