
In real-world transport problems, the occurrence of data uncertainty is a common and natural phenomenon. Uncertainty in traffic data stems from the fact that it is impossible to precisely determine data values at the planning stage of a transport process, e.g., when routing vehicles. Accurate real-world values are always obtained after the fact, meaning they automatically become historical data.
Using historical data to plan future real-world processes will always be burdened with uncertainty, as there is no hundred-percent certainty that a change in the value of a given parameter will follow the exact same course as it did previously. Therefore, data uncertainty results from the nature and characteristics of the described phenomenon and implies the impossibility of establishing exact future data values during the decision-making process. In routing problems, the primary data consists of travel time information, which is highly sensitive to changes in its values.
Failure to account for data uncertainty at the planning stage can result in generating a solution that incurs unnecessary costs during actual implementation. Work [7] presents an analysis of linear programming problems from the NETLIB database, which shows that even a small change in input data values – 0.01% – caused constraint violations of up to 210%. A constraint for vehicle routing problems could be, for example, a vehicle arriving at a customer’s location within the appropriate time window. An example of the impact of model parameter fluctuations is presented in work [2], which analyzed the Vehicle Routing Problem with Time Windows. A 1% change in travel time caused the resulting solution to change, which consequently altered the sequence of visited customers. This fact is certainly significant from the perspective of the enterprise, its transport implementation costs, as well as from the perspective of customer service levels. The assumption that available urban traffic data is one hundred percent certain is a false assumption with no justification in reality.
One solution for dealing with data uncertainty can be the theory of robust optimization. The following article presents a general characterization of this field and its current application in the Vehicle Routing Problem (VRP). As it turns out, based on a literature review, robust optimization is a current trend in VRP research.
Source of traffic data uncertainty
As already mentioned, uncertainty in data regarding processes and objects occurring in a city’s transport system is a natural phenomenon. This results from the basic features and properties of a city’s transport systems, which are (based on [1]):
– a large-scale system consisting of many objects and subsystems between which relationships and feedback loops occur,
– the presence of many classes of users whose interactions and communication behaviors are often based on the psychophysical state of the users,
– the occurrence of complex traffic phenomena characterized by randomness and indeterminacy, e.g., traffic incidents,
– high dynamism of phenomena, e.g., the formation of traffic jams,
– susceptibility to external factors, e.g., weather conditions significantly change driver behavior, which in turn visibly affects the fluidity and intensity of traffic.
In order to properly plan transport processes, i.e., to determine routes for vehicles that will perform cargo service for a given city area, it is necessary to include the aforementioned properties in the decision-making process. The factors cited above can influence changes in travel times; therefore, assuming in the planning process that the data possessed is accurate and certain is incorrect and does not correspond to reality.
The article contains 36,280 characters.
Source: Logistyka Magazine








