Estudio e implementación de algoritmos de aprendizaje automatizado para el pronóstico de la actividad pluvial en Bogotá

Abstract

This work focuses on the study and evaluation of different automated learning methods applied to the prediction of precipitation with information from weather stations located in the city of Bogota, Colombia. Short-term weather forecasting is an extremely complex field that has been the subject of exhaustive study in recent years. Advances in measurement and data capture systems, as well as in computational capacity, have made possible a remarkable improvement in the accuracy of short-range weather forecasts. This progress is having a significant impact on the way individuals and communities anticipate and prepare for weather conditions. The automated learning methods selected for this study encompass both fuzzy systems and neural networks. Among the fuzzy systems, simple genetic algorithms (ANFIS), evolutionary algorithms (DE) and fuzzy system random search algorithms (random search) were implemented. On the other hand, in the field of neural networks, FeedForward networks, LSTM networks and BILSTM networks were used. All the implementations were carried out using algorithms developed in the MATLAB programming environment. The evaluation of the implemented methods was carried out by analyzing various performance indices, such as prediction of change in direction (POCID), pearson correlation coefficient (r), nash-sutcliffe efficiency (NSE) and root-mean-square deviation (RMSE). The results obtained reveal significant findings that underscore the need for further research in this area. This paper presents a solid starting point for understanding and improving precipitation prediction methods in Bogotá, thus contributing to the advancement of research in climatology and machine learning applied to unpredictable and complex meteorological phenomena.