A commodity-based production and distribution road freight model with application to urban and regional New South Wales

Abstract

The complexity of freight demand forecasting coupled with non-availability of data of the required scale and features often limits its inclusion in demand forecasting. Available data on many of these aspects of freight, at varying degrees of aggregation spatially, are publicly available for modelling in Australia. This paper provides a novel approach based on the principle of entropy maximisation to combine these diverse datasets to develop the commodity-based production and distribution first component of a freight behavioural logit model for New South Wales (NSW). The implementation of the model is presented using NSW as case study. Key outcomes include the ability of the model to reproduce several observed aggregate results including the sum of all commodities produced and/or consumed in each region of NSW; the average distance (kilometres) covered by each vehicle class in the study area and the average distance each commodity group is transported across the study area.

KEYWORDS:

• Freight models
• aggregate data
• commodity flows
• commodity generation model
• logit share model
• MetroScan
• New South Wales

Acknowledgements

This paper contributes to the research programme of the Volvo Research and Education Foundation Bus Rapid Transit Centre of Excellence. We acknowledge the Foundation for partial funding support and the detailed comments of three referees and William Lam.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Our proposed model is aggregate based and not agent based as data required for modelling at this level of disaggregation is not available in our study area. The promising approach adopted is to group businesses or agents into useful and manageable zones, where each commodity can be seen as been produced from a zone and/or consumed at various zones. The production and consumption models as shown in Figure 2 below are transformed into Nested Logit models (linked logit), which are known to be behavioural (Ben-Akiva and Lerman 1985; Hensher, Rose, and Greene 2015; Lam and Lo 1991). Under the logit framework, the decision-makers do not necessarily have to be agents or individuals (see Train 2009). For example, Equation (26) below represents the probability of producing each commodity group in each production zone and distributing them among various consumption zones by each vehicle class.

2 Note that this formulation is still valid even if different states of the system of interest are specified.

3 We have a traffic assignment model for the study area that provided the distance data for each vehicle class. If one does not have a suitable traffic assignment or network model, one can use ‘crow fly’ distance or the Manhattan distance which require only the zonal coordinates (e.g., latitude & longitude) to compute. Alternatively, one can ignore that constraint as explained above. In practice, network data like distance or time are best derived from a traffic assignment model.

4 MetroScan is an urban model system but for freight we recognise that it moves on trucks and trains that often begin and /or end their trips outside of the urban area. This is taken into account with external O and D zones so as to capture the full freight task within the urban area.

5 The process of validating a model is outside the scope of this paper. The basic principle underlying the proposed model is that ‘every data point explains certain aspect of the freight problem’ so setting having a hold out sample as suggested by a referee (from data which is already very limited) weakens the explanatory power of the model. In practice, model validation (complex iterative process) is done at the assignment stage where assigned vehicle flows are compared with observed vehicle counts. This process is undertaken within the MetroScan system for NSW that incorporates other models like the passenger transport models as they share the same road space. The models developed in the current paper feed into the transport components of MetroScan or indeed any integrated transport and land use modelling system. We should add that the issue of setting some data aside for validation is no longer enforced or encouraged in practice. All Transport for London (TfL) models do not set data aside for validation as data collection is very expensive to collect the minimum data required for calibration purposes.

6 Each polygon in Figure 3 represents a zone where each commodity can be produced or consumed. The zone is equivalent to NSW’s Statistical Area Level 2 (SA2), which is designed to represent a community that interacts together socially and economically (Australian Bureau of Statistics 2018).

Additional information

Funding

This work was supported by Volvo Research and Educational Foundation Bus Rapid Transit Centre of Excellence.