A significant growth of the world’s population together with fast growing urbanization is causing challenges for cities in the future. One of these challenges is how to deal with waste production and therefore the waste collection in such areas. Smart waste collection is a promising solution since it optimizes an already excising infra structure including vehicles and since the collection and transportation of the produced waste accounts for roughly 70% of the waste management costs. Smart waste collection is a routing problem and can be categorized both as a vehicle routing problem (VRP) or as an inventory routing problem (IRP) depending on which container selection method is used. Smart waste collection uses sensors to measure and communicate the fill levels of the waste containers. This data can be used to optimize the process of waste collection and optimize the chosen KPIs. When all fill levels are known, routes can be optimized and containers can be collected at exactly the right time. This article will use real data obtained from the containers collected by OMRIN with the help of AMCS and their software. Container selection methods can be categorized as threshold based, attractiveness based or must-go may-go based. Containers can be selected for collection, roughly based on three methods. The interesting thing is that these three methods have not yet been compared to each other. Therefore the main goal of this research is to compare the three aforementioned methods based on real data and investigate certain tuning parameters to optimize each model based on the chosen KPIs. The models are first individually optimized by changing their tuning parameters and keeping the overflow in an acceptable range of the baseline, calculated based on real data. Overall a strong negative correlation is found between the total traveled time and the amount of overflows. Furthermore a warm-up period of three days is used, meaning the first three days of a test instance will be removed in order to capture only the steady state of the models. The individual optimization shows the best solutions for a 1% threshold buffer for the threshold model, a three-day horizon with 110% upper limit for the attractiveness model and a three-day horizon with an threshold buffer of 7% and a 110% upper limit for the must-go may-go model. When compared the attractiveness model shows to have the smallest total travel time. However, its computational time exceeds the actual total travel time. The must-go may-go model shows a slightly higher total travel time, but has a significant lower computational time. Therefore making it more suitable for real world application. This research as well shows that the two-day horizon instances of both the attractiveness model and the must-go may-go model barely improve the solution of the one-day horizon. Finally a true forecasting model was used to see what potential lies with designing a detailed forecasting model, which shows to be in the same order of improvement as was obtained by tuning parameters of each model.

The petrol distribution problem is well-known in the literature as the Petrol Station Replenishment Problem (PSRP). This research concerns a "rich" version of the multi-period PSRP with real-life characteristics, to represent the complexity of the practical problem. The planning is optimised over multiple days, since stations do not need to be replenished every day. Mixed-Integer Linear Programming (MILP) models and a decomposition heuristic are proposed as planning methods, which are evaluated with a case study based on a real-life petrol distributor. Variants to these MILP models are proposed for the situations where the inventory is allowed to drop below the safety stock level, where inventory levels need to be minimised and where the service time depends on the delivery quantity. Moreover, to determine delivery quantities, the heuristic uses the new introduced simultaneous dry run inventory policy. An improvement procedure is applied to improve the initial heuristic solution. A commercial solver is able to find solutions for instances with up to 20 stations and 7 days for the MILP models. Exact solutions are found for instances up to 10 stations and 5 days. A heuristic solution was found for the full case study of 59 stations and 14 days, within the time limit of two hours.