Análise comparativa da aplicação de modelos para imputação do volume médio diário de séries históricas de volume de tráfego

Abstract

In order to improve the road system, with regard to its infrastructure and operation, it is necessary to perform studies and planning, by seeking the best use of existing resources. Therefore an important traffic measure is used, i.e., vehicle volume. Traffic data is collected either manually or electronically; however both ways can fail and not collect all data. In the case of electronic counting equipment, the continuous data collection may form a time series, which produces failures in the database due to non-collection, which can compromise the studies based on this information. Therefore this work aims to perform analysis of methods used to estimate these missing values, by trying to know the most effective model for the Average Daily Volume variable of the data obtained by the continuous counting stations installed in the state highways of Ceará. The estimation models used in this work are the ARIMA models for time series analysis, and simple models, which present a less complex application and a faster processing, while the ARIMA requires more specific knowledge of the professional who uses it. The most effective method considered herein was the one that obtained smaller errors after the application of the models. Four permanent counting stations were selected for these applications, according to the percentage of valid data and its location, by seeking the use of stations in representative points of the state. The best model found was ARIMA (1,0,1)7 (with an average error of 1.816%), however one of the simplest models, MS2, produced results similar to those of ARIMA (an average error of 1.837%), and it can also be considered suitable for application in the allocation of missing values.