Resource Constrained Project Scheduling: a Hybrid Neural Approach

This study proposes, develops and tests a hybrid neural approach (HNA) for the resource constrained project scheduling problem. The approach is a hybrid of the adaptive-learning approach (ALA) for serial schedule generation and the augmented neural network (AugNN) approach for parallel schedule generation. Both these approaches are based on the principles of neural networks and are very different from Hopfield networks. In the ALA approach, weighted processing times are used instead of the original processing times and a learning approach is used to adjust weights. In the AugNN approach, traditional neural networks are augmented in a manner that allows embedding of domain and problem-specific knowledge. The network architecture is problem specific and a set of complex neural functions are used to (i) capture the constraints of the problem and (ii) apply a priority rule-based heuristic. We further show how forward-backward improvement can be integrated within the HNA framework to improve results. We empirically test our approach on benchmark problems of size J30, J60 and J120 from PSPLIB. Our results are extremely competitive with existing techniques such as genetic algorithms, simulated annealing, tabu search and sampling.