Estimating freight origin destination matrices using combined commodity flow survey and roadside survey data

Abstract

The paucity of available data was limiting studies of freight flow in Thailand. To overcome this problem, commodity flow survey (CFS) and comprehensive freight transportation by truck using roadside survey (RS) were launched to collect comprehensive freight flow data throughout the kingdom of Thailand. Since these two surveys were pioneering and due to budgetary limitations, the resulting data are still incomplete and must be adjusted. The need to produce a freight origin destination matrix using available data from CFS and RS led to the objectives of this research. This research has two main objectives. The first is to develop a methodology for combining CFS and RS. The second is to develop a method for filling gaps in the origin destination matrix based on the Adaptive Neuro Fuzzy Inference System (ANFIS) approach. The methodology to combine these two data sources was developed which uses the strengths of each method, the CFS distribution pattern and the RS marginal total. The first method is Trip Length Distribution Adjusting (TLDA), which uses adjustments to CFS trip length distribution to meet RS marginal total. The second method is Gravity Model Approach (GMA), which uses CFS friction functions to adjust RS data matrix. The method was calibrated using two difference sources of roadside survey. The results indicated that the adjusted volumes of the two data sources agreed despite being collected at different times and by different authors, and that the differences between the total adjusted volumes were quite small. It can therefore be concluded that the developed method can be used to adjust the data. For the second component, a model using BOX-COX transformation and Adaptive Neuro Fuzzy Inference system (ANFIS) was developed and verified against a convention gravity model. Two types of model, using convention gravity variables and using socio-economic variables, were developed. The results showed that the ANFIS model outperformed both the conventional gravity model and the BOX-COX model. These results proved the performance of the adaptive neuro fuzzy inference system for modeling complex system and its ability to model freight trip distribution.