Vehicle Routing by Learning from Historical Solutions

Abstract

The goal of this paper is to investigate a decision support system for vehicle routing, where the routing engine learns from the subjective decisions that human planners have made in the past, rather than optimizing a distance-based objective criterion. This is an alternative to the practice of formulating a custom VRP for every company with its own routing requirements. Instead, we assume the presence of past vehicle routing solutions over similar sets of customers, and learn to make similar choices. The approach is based on the concept of learning a first-order Markov model, which corresponds to a probabilistic transition matrix, rather than a deterministic distance matrix. This nevertheless allows us to use existing arc routing VRP software in creating the actual route plans. For the learning, we explore different schemes to construct the probabilistic transition matrix. Our results on a use-case with a small transportation company show that our method is able to generate results that are close to the manually created solutions, without needing to characterize all constraints and sub-objectives explicitly. Even in the case of changes in the client sets, our method is able to find solutions that are closer to the actual route plans than when using distances, and hence, solutions that would require fewer manual changes to transform into the actual route plan.

Probability estimation and structured output prediction for learning preferences in last mile delivery

“…Both of the above-mentioned studies focus on point to point commute for individual drivers. Canoy & Guns (2019) approach the problem in a vehicle routing setting from a different perspectivethey introduce a weighted Markov model to learn the preferences from historical routes. This technique avoids the need to explicitly specify the preference constraints and implicit sub-objectives.…”

“…Learning zone order from historical data. We adapt the approach introduced in Canoy & Guns (2019) of learning transition probabilities between zones from the historical data. This approach models preferences as a Markov chain, where the preferences are learnt as the probability of going from one zone/stop to another.…”

“…We take inspiration from Canoy & Guns (2019) where experimental results show that a cost matrix that mixes (1) distances and (2) transition probabilities, can lead to better solutions being found. From an application point of view, this is motivated by the observation that planners have personal and experiencebased preferences, but also always take the total distance into account.…”

“…We first assume that preferences can be captured as costs of going from one zone/stop to another zone/stop. This assumption is standard as it is used in Canoy & Guns (2019); Mandi et al (2021); Chen et al (2021). One of the advantages of this assumption is the possibility to use exact and/or approximate algorithms available for the TSP.…”

Learn-n-Route: Learning implicit preferences for vehicle routing

“…Our proposed first order Markov model approach has been published in a conference proceeding [Canoy and Guns 2019] and in this article, we extend this methodology by developing the following new contributions:…”