GPU-accelerated Logistics Optimisation for Biomass Production with Multiple Simultaneous Harvesters Tours, Fields and Plants

Abstract: Within the context of biomass production, this paper proposes a method for logistics optimisation. Starting from a headquarter multiple tours are to be executed simultaneously by groups of harvesting units (HUs) and support units (SUs) to first harvest biomass from multiple agricultural fields, before supplying the biomass to multiple biogas plants (BPs) via shuttling SUs. This problem is relevant on a larger scale in particular for contractors. This problem is complex since there are three interconnected optimisation levels: (i) the assignment of BPs to tours and the ordering of BPs assigned to each tour, (ii) the assignment of fields to BPs and the ordering of fields assigned to each BP, and (iii) determining the number of HUs and SUs assigned to each tour, whereby different HUs and SUs may in general have different working rates and loading capacities. Problem modeling and a solution method are discussed. For the latter, a GPU-accelerated heuristic search algorithm is proposed. For the former, an optimisation criterion minimising both total accumulated path length and the maximum completion time over all harvesters tours, an embedded local minimisation for the assignment of SUs to tours, and demand fulfilment constraints for BPs are discussed. In stochastic simulation experiments it is found that permitting unconstrained assignment of any available field that a contractor services to any available BP, independent of their ownerships, is crucial (i) to attain maximum path length savings, and (ii) also to best balance and minimise uniform completion times over all harvesters tours such that weather-dependent harvesting windows can be exploited optimally.