Diseño de un pronóstico para optimizar visibilidad y gestión de contenedores refrigerados vacíos

Abstract

This project aims to develop a forecasting algorithm to improve the visibility and operational management of empty refrigerated containers in a shipping line, increasing the accuracy of weekly forecasts and reducing stockouts through reliable information. The hypothesis is that centralizing historical export data with validated logistics variables improves operational planning, compared to an environment with limited visibility, scattered information, and high demand variability. The development included the collection of nine critical variables from corporate databases and operational reports from terminals and warehouses. The reliability and validity of the variables were evaluated using statistical tests and 1:1 comparisons between sources. It was determined that the weekly contract variable was not valid and was replaced by actual export volume. Models such as SES, Holt-Winters, and ARIMA were also analyzed. The results confirmed the validity of the variables. The SES model performed best with MAPE = 15.72% and RMSE = 348.3. The design achieved 100% of the weeks evaluated without stockouts, 71% process automation, 100% compliance in overstay control, and an estimated 45.94% reduction in CO2 emissions, concluding that a forecast based on historical evidence improves logistics efficiency. Keywords: Forecasting, Data validation, Refrigerated containers, Operational logistics, Stock outs.