Implementación de un sistema prototipo para llevar a cabo pronósticos meteorológicos a corto plazo en las zonas de Cumbayá, Calderón y Conocoto, basado en una red neuronal.

Abstract

In the city of Quito, not all the territory presents the same meteorological conditions, therefore, the study of microclimates is an important aspect to consider for the precise prediction of the climate, however, the installation of equipment or meteorological stations in various areas. The purpose of this project is to implement a prototype system to carry out short-term meteorological forecasts of temperature and relative humidity in the areas of Cumbayá, Calderón and Conocoto, based on the use of a neural network; whose input information will be obtained from low-cost automatic weather stations (AWS). Currently, there are several hardware and software tools available on the market with very good performance and low cost for the design and implementation of a robust and efficient AWS, which allows continuous monitoring of meteorological data, to obtain a history of climate behavior in the different areas, information that will allow the development of a shortterm weather forecast model based on the use of the ARIMA prediction model and neural networks such as: simple LSTM and stacked LSTM (2 layers); in order to obtain the model with the highest performance, precision and reliability in temperature and relative humidity forecasts. Finally, the lowest errors were obtained with the simple LSTM neural network, determining that the best predictions occur in localities with greater climatic stability and continuity in data acquisition.