Neural Networks in the prediction of microclimate, La Hechicera-Mérida-Venezuela, Case of Study

Abstract

An Artificial Neural Network (RNA) was developed with capacity to make short-term climatic predictions, trained with data at intervals of 15 min for 454 days in a climatological station located in La Hechicera Mérida-Venezuela (1896 masl). For training and validation, were used simulated patterns that contain daily variations of radiance and temperature. Three test functions were chosen such as the Sine, Cosine and the Legendre Polynomials Pl(x). The (RNA) e.g, can predict the interval (0,1] using only as input the interval [-1,0] of the corresponding derivative P´l (x)  of the polynomial. In the production phase with real data, it was found that the network is able to predict the temperature with approx. 5% error in the hourly range [12:15 to 06:15] pm, only with the temperature data in the range [6:00 to 12:00) am. Also, when predicting temperature from radiance (5%), radiance-radiance (16%). The use of (RNA) for the prediction of micro-climate in the short term is considered feasible, being able to extend its use to other localities, which could be useful for the development of disaster prevention plans, sowing periods, prediction of energy supply in wind and solar power stations.