Optimal sensors placement for waste collection

Waste collection is a sensible topic in the context of smart cities. The tactical collection operation usually leaves room for improvement mainly in the form of schedule optimisation. The study in question explores the benefits of having access to additional contextual information, specifically data related to the dumpsters’ filling process. In this setting, there are two ways to predict the current waste volume held by each of the dumpsters: the first, which introduces high variance and can only be performed during the collection phase, relies on an estimation through weighting performed by the garbage truck; the second, which provides higher precision, utilises volumetric sensors positioned inside the dumpsters. Having low uncertainty incentivises more fruitful and less frequent trips, while reducing the risk of exceeding maximum capacity. The choice of how many sensors to employ and where to place them, therefore, must focus on maximising the cost reduction of the collection procedure and guarantee that it compensates for the installation and maintenance cost of the sensors themselves. The goal of the study consists in finding the optimal placement of volumetric sensors inside a predefined set of dumpsters scattered around the Metropolitan City of Turin in order to minimise the total operative cost of a fleet of vehicles predisposed for the collection of waste. As the filling rate of every dumpster is modelled as a random variable, this is a stochastic programming problem. From a mathematical point of view, a heuristic was developed to optimise the sensors’ location, relying on an Adaptive Large Neighbourhood Search algorithm for scheduling the set of dumpsters to visit each day and on Travelling Salesman Problem solvers for optimising their order of visit. Each day’s solution also takes into account following days’ schedules in order to meet the current requirements without compromising the ability to achieve the same results in the future.