Computational Intelligence Approaches for Optimising Complex Production Scheduling Problems in the Nigerian Petroleum Services Industry

This dissertation develops and validates novel computational intelligence approaches for optimising complex production scheduling problems encountered in the Nigerian petroleum services industry, addressing both the theoretical gap in metaheuristic development for industry-specific scheduling variants and the practical need for decision support tools in a sector of critical national economic importance. Petroleum services scheduling involves multi-objective job sequencing with stochastic processing times, resource-sharing constraints across multiple project sites in the Niger Delta, and dynamic job arrival patterns that render exact optimisation approaches computationally intractable at operational scale. The research contributes an original hybrid optimisation algorithm integrating an improved adaptive large neighbourhood search with a deep reinforcement learning component trained to dynamically select destruction and repair operators based on solution space exploration history. The algorithm is developed, calibrated, and tested on benchmark scheduling problem instances before application to real petroleum services scheduling datasets provided by three operating companies in Rivers and Delta states. Computational experiments demonstrate that the proposed algorithm outperforms published best solutions on twelve of fifteen standard benchmark instances and achieves schedule cost reductions of eighteen-point-seven percent on average across the industrial test cases compared to current company scheduling practice. Keywords: computational intelligence, production scheduling, metaheuristic, deep reinforcement learning, petroleum services Nigeria