Identificação de variáveis prevalentes para situações de stress em parques de tancagem: uma análise a partir das redes neurais artificiais

Abstract

The improvement in operations planning is a constant concern of oil refineries, and the efficient management of inventories in tank farm sites is a key point in this context. However, there are few studies that deal specifically with this issue and existing simulation tools are expensive and do not reflect the reality of many refineries. The inefficient management or improper sizing of tank farm sites, in turn, can generate significant financial losses. A oversized tank farm generates management and maintenance costs, in addition to inventory costs. An undersized tank farm can generate losses due to shortages and degradation of products, among others. The situations in which the undersizing generates impact in operations, with or without financial losses, is defined as stress in tank farm sites. The present study describes the implementation of a tool that allows the quantification of stress in tank farm sites. This measurement is made using a heuristic based on data on the inventory movimentation, maintenance status and inventory levels of the tanks. It also describes the calculation method of fifty-nine variables related to the movimentation of inventory. The influence of these variables on the formation of stress situations was analyzed using artificial neural networks. This influence was quantified in five different scenarios, considering whether or not a product certification cycle and the nature of the variables. As a result, it is identified that the variables related to emptying time, tank downtime at low level and fill time are prevalent in creating stressful situations in tank farm sites in the analyzed context. They are also mapped and formalized flows of algorithms to determine the stages of a tank cycle, and is defined a formula for determining the stress level in a tankage park at a given time.