Modelos lineales generalizados en el análisis de derivados importados y refinados en la empresa Pública Petroecuador en el periodo 2017-2020.

Abstract

The analysis of refined and imported derivatives through pipelines in the Petroecuador Public Company, aims to forecast the number of barrels of derivatives that are distributed at the national level and the appropriate pipeline, considering changes in the factors on which they depend, using generalized linear models (GLM). For the study, 224 observations recorded in the 2017-2020 period were considered, with information on the volume transported in barrels of each year, in each pipeline, through the different sections, as well as the product that was distributed, in addition, the flow, capacity and volume dispatched. With this information, the quantitative response variable of interest was proposed, which is the transported volume, and a multiple linear model was adjusted. The performance of the model was evaluated; multivariate outliers were detected; fit was improved by selection of significant predictor variables. A generalized linear model was applied, which is the Poisson regression; the model was evaluated; and, to select the best model from those proposed, two criteria were considered: the comparison of models based on the sum of squares and the Akaike information criterion. For the qualitative multi-pipeline response variable, a logistic regression model was adjusted. After detecting 87 atypical data, with multiple regression, a coefficient of determination of 92% was obtained, it meets two of the assumptions and fails in independence; while the Poisson regression presented over-dispersion. Whereas, for the logistic regression it will depend on the transported volume and the product. It is concluded that multiple egress is the most appropriate to identify the transported volume, and nominal logistic regression to identify the appropriate pipeline. Finally, an adequate decision making by the company's staff is recommended in the face of the different scenarios that may arise.