Browsing by Author "Eygelaar, Jancke"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
- ItemGenerator maintenance scheduling based on the expected capability of satisfying energy demand(Stellenbosch : Stellebosch University, 2018-03) Eygelaar, Jancke; Van Vuuren, J. H.; Stellenbosch University. Faculty of Engineering. Dept. of Industrial Engineering.ENGLISH ABSTRACT: The integrity of an electric power system is significantly threatened by unexpected downtimes of power generating units (PGUs). In order to minimise the occurrence of such unexpected PGU downtimes, planned preventative maintenance is routinely performed on the PGUs of the system. The effective scheduling of these planned maintenance PGU outages is a considerable challenge for any power utility. The celebrated generator maintenance scheduling (GMS) problem involves finding a set of planned preventative maintenance outages of PGUs in a power system. A feasible solution to this problem is typically a list of dates indicating the commencement of planned maintenance for each PGU in the power system. Solutions to the GMS problem are typically subjected to a wide variety of power system constraints and the problem is considered to be a hard combinatorial optimisation problem. Two novel GMS criteria are introduced in this dissertation. The first criterion involves minimisation of the probability that any PGU in the system will fail during a scheduling window of pre-specified length. This scheduling criterion is weighted according to the rated capacity of each PGU so as to give some preference, in terms of maintenance commencement times, to PGUs that contribute considerably to the overall system capacity. This criterion draws from basic notions in reliability theory which may be used to estimate the failure probability of a system. The second criterion involves maximisation of the expected energy produced during the scheduling window. In this case, PGU failures are modelled by random variables. Two mixed integer programming models are formulated for the GMS problem with the newly proposed scheduling criteria as objective functions. One of these models is linear and the other one is nonlinear. The nonlinear model is linearised by piecewise linear approximation in order to be able to solve it exactly. Both models incorporate a number of constraints, including energy demand satisfaction constraints, earliest and latest maintenance window constraints, maintenance resource constraints and maintenance exclusion constraints. Two GMS test problems are modelled in this fashion. The resulting four GMS model instances are each solved by two different solution approaches - exactly and approximately (by employing a metaheuristic). The exact solution approach involves use of IBM ILOG's well-known optimisation suite CPLEX which employs a branch-and-cut method, while the metaheuristic of simulated annealing is implemented in the programming language R as approximate model solution methodology. After computing optimal solutions for the four GMS model instances mentioned above, a sensitivity analysis is performed in order to determine the feasibility of an exact solution approach in respect of small to medium-sized GMS problem instances. An extensive parameter optimisation experiment is also conducted in order to obtain a suitable set of simulated annealing parameter values for use in the context of the four GMS model instances. The GMS solutions obtained when incorporating these suitable parameter values within the simulated annealing algorithm for the four GMS model instances are compared to the corresponding exact solutions. The approximate solution methodology is found to be a viable alternative solution approach to the exact solution approach, capable of obtaining solutions within 3% of optimality for all four GMS model instances | often requiring considerably shorter computation times. The e cacy of the two proposed scheduling criteria are also analysed in terms of a real-world case study based on the power grid of the national power utility in South Africa. The 157-unit Eskom test problem contains a large number of PGUs, some of which require maintenance multiple times during the scheduling window. The aforementioned approximate solution approach is adopted to solve this large problem instance with respect to both proposed scheduling criteria and is found to be a viable approach from a practical point of view. A computerised decision support system (DSS) is also proposed which is aimed at facilitating effective GMS decision making with respect to the two proposed scheduling criteria. The DSS is implemented in the programming language Shiny, which is an R package for creating user-friendly interfaces. The DSS is equipped with an intuitive graphical user interface and the system is able to solve user-provided problem instances of the GMS problem.