Neural Network Architecture for Scheduling Tank Trucks Loading at Petroleum Products Storages

This study presents a decomposition of the scheduling problem for the process of petroleum product dispatch from petroleum storage facilities using road transport into major functional blocks. The proposed decomposition is aimed at developing a specialized modular architecture of an artificial neural network to address the scheduling problem for petroleum product dispatch via road transport, as well as for solving scheduling problems of similar classes. It is demonstrated that the neural network task of assigning work groups to executor groups can be reduced to a sequence-to-sequence transformation problem utilizing attention mechanisms. Neural architectures implementing the identified functional blocks were selected, along with appropriate methods for training the model. The resulting modular components enable separate, block-wise training of target models for constructing suboptimal schedules.