Conference Agenda

Waste collection management has seen an increasing interest in the OR community these past years. This is due to the wide range of Capacitated Vehicle Routing Problems (CVRP) studied and the growing trend in studying sustainability-related problems.

For waste collection companies, it can be challenging to identify the type of strategy most suitable for a given situation. On the one hand, the complexity of the state-of-the-art algorithms presented in the literature; on the other hand, the data needed for these algorithms can be challenging to obtain and encode.

In partnership with Alpenluft, a Swiss waste collection consulting company, and the Innosuisse agency supporting R&D projects, we developed the WasteLogs application, a user-friendly strategic waste collection decision tool. The application offers interfaces allowing the encoding of the collection points, the amounts of waste to collect, and the collection strategy in different features that can be combined to generate a routing for collection vehicles. There are currently three state-of-the-art collection strategies implemented in the tool. Each algorithm minimizes the CO2 emissions through heuristic methods; the user can then identify what collection strategy is the most suitable and extract the information needed to import them into GPS systems. WasteLogs also allows importing existing collection tours to evaluate whether they can be improved.

We consider a decision problem in a transport company. Every day, the company receives transport requests by phone from various customers. A request consists of an origin, a destination and a time window within which the request must be fulfilled. The decision to accept or reject the request must be made within a short timeframe: between a few seconds and a few minutes.

We develop an analytical decision process to accept or reject a request based on expected revenues. We use a geographic information system to calculate transport times and lengths. We use the company's historical data to predict new spatio-temporal requests, using clustering and probability methods. Next, a simulation generates instances of vehicle routing problems, which are solved using heuristics from open-source tools, enabling the expected revenue to be calculated. The request is then accepted or rejected, based on the expected value at the time the request is made. We test our methods on generated data.

The maintenance of an electricity distribution network involves numerous daily technical interventions. In this problem, we are given a set of interventions each with associated time windows, location, necessary skills and duration, as well as a set of teams of technicians with associated set of skills, each represented by a vehicle. We need to find feasible routes on the interventions for each vehicle, taking into account the time windows and skills, and ensure that each vehicle returns to its departure depot before the end of the day. The primary objective is to maximize the total duration of completed interventions and as a secondary objective, we aim to minimize the overall routing cost. This problem can be formulated as a capacitated vehicle routing problem with time windows. Due to the large number of vehicles and interventions, this results in a large-scale optimization problem, and its operational nature limits the time available for exact solving. Here, we propose a column generation approach where the subproblem decomposes into a subproblem per vehicle and each potential route of a vehicle is considered as a new column in the master problem. To generate these routes, we rely on dynamic programming. Real-world instances from EDF (Electricité de France) of historical technicians' interventions will be used to evaluate the effectiveness of the proposed methods.