Análise de métodos de previsão para demandas Intermitentes de itens de uma empresa de transporte ferroviário

Abstract

For companies that perform maintenance on their assets, it is very important that when one part fails, there is another available in stock for the exchange. However, it is difficult to know when any item will fail, and generating parts inventories to prevent this problem can cause unnecessary expenses to the company. As a result, these companies have been investing more and more in inventory management, in order to reduce their capital stock, without affecting the availability of material for application. In order to have good inventory management, the use of time series for forecasting demand has been gaining space in companies. In addition to getting accurate forecasts, companies need to rank their items in groups that have common characteristics. Separating the items according to their budgetary representativeness for the company, and the periodicity and variability of the quantity consumed, helps to categorize them adequately. There is a group of items that is particularly difficult to predict, because their components have few periods with demand and there is great variation in the amount consumed; this group is said to have Intermittent demand. In this study, we analyze the demand for items, exchanged in corrective maintenance by a railway company, which have intermittent demand characteristics. This group of items was subdivided according to their budget importance for the company, generating three subgroups. We calculated the forecasts using the Simple Average method used by the company, and also by Moving Averages, Exponential Smoothing, the Croston Model and the Adjusted Croston Model. We also calculated the Mean Absolute, Mean Quadratic and Mean Quadratic Geometric errors, and analyzed them. We concluded that the models of Croston and Croston Adjusted were the ones that presented the smallest errors, and are therefore the ones indicated for application in the company, instead of the model now in use.