Freight Activity Chain Generation Using Complex Networks of Connectivity

Freight vehicle movement can be captured using activity chains, which capture the freight related activities both spatially and temporally. This is an advancement over origin-destination matrices, since freight vehicles are studied at a disaggregate level. Digicore Fleet Management has provided GPS data of over 40,000 freight vehicles travelling in South Africa over a six month period, from which freight activity chains can be extracted. The first contribution of this paper is to generate synthetic freight activity chains using a complex network that captures the connectivity between firms where freight activities take place, and therefore also captures some aspect of the behaviour of the vehicles. The complex network is built with the activity chains of 60% randomly selected freight vehicles, called the training set. 10 synthetic populations are generated using this complex network, each representing a 10% sample of the total freight population in South Africa. As a second contribution, we use the observed activity chains of the remaining 40% of vehicles, called the test set, to compare to the synthesised activity chains. The results indicate that the complex network approach of generating synthetic populations correctly estimates the fraction of activity chains for different chain lengths. The chain start times of the synthetic populations are generally similar to that of the observed activity chains, except for some underestimation during the morning peak hours. Vehicle kilometres travelled are only slightly overestimated in the synthetic activity chains. This complex network approach to generating synthetic freight populations is novel, and the results indicate that the synthetic populations generated are an accurate representation of South Africa's freight population.