Download files
Abstract
As energy prices continue to rise, enterprising manufacturers are incorporating self-managed microgrids to reduce energy costs. This comes with the task of scheduling various energy supply assets, a non-trivial undertaking in itself. However, to achieve near optimal energy cost minimisation, it is also necessary to modify production scheduling. Scheduling both energy supply and production processes is a large problem area about which comparatively little research has been conducted.
This thesis bridges the aforementioned gap, by firstly designing a conceptual model for an energy resource and production scheduling system. Then a solution methodology comprising a hybrid algorithm, broadly based on the Fast Non-Dominated Sorting Genetic Algorithm (NSGA-II) structure with an inner Mixed Integer Linear Programming (MILP) energy optimisation procedure, is developed to efficiently solve the multi-objective problem of minimising energy cost and optimising a production metric. Various crossover and mutation functions are extracted from literature and tested in a full two-factor experiment. The best combination of operators are chosen based on their performance identified through a robust statistical analysis. The system's response surface to various operator settings is modelled and optimal parameter values are identified via a Particle Swarm Optimisation (PSO) procedure. A novel repair operator is also proposed, itself a linear MILP optimisation procedure, which is found to drastically improve the performance of the overall system. Techniques such as a variable Mixed Integer Programming (MIP) gap stopping criterion were utilised in the energy cost optimisation sub-model to improve solution times.
Two case studies, each containing multiple analyses of various microgrid asset combinations, are used to compare the proposed system to a previously developed single-objective MILP method. While the proposed system is proficient in optimising the production metric, it excels in its primary purpose of energy cost minimisation. In most cases the proposed system is able to find solutions that equal or surpass the MILP system in regard to energy cost. In addition to higher quality edge solutions, it also provides the operator with a clear view of the inherent trade-off between the two objectives and represents a more computationally efficient approach.