Scheduling coal handling processes using metaheuristics

Abstract

The operational scheduling at coal handling facilities is of the utmost importance to ensure that the coal consuming processes are supplied with a constant feed of good quality coal. Although the Sasol Coal Handling Facility (CHF) were not designed to perform coal blending during the coal handling process, CHF has to blend the different sources to ensure that the quality of the feed supplied is of a stable nature. As a result, the operation of the plant has become an extremely complex process. Consequently, human intelligence is no longer sufficient to perform coal handling scheduling and therefore a scheduling model is required to ensure optimal plant operation and optimal downstream process performance. After various attempts to solve the scheduling model optimally, i.e. with exact solution methods, it was found that it is not possible to accurately model the complexities of CHF in such a way that the currently available exact solvers can solve it in an acceptable operational time. Various alternative solution approaches are compared, in terms of solution quality and execution speed, using a simplified version of the CHF scheduling problem. This investigation indicates that the Simulated Annealing (SA) metaheuristic is the most efficient solution method to provide approximate solutions. The metaheuristic solution approach allows one to model the typical sequential thoughts of a control room operator and sequential operating procedures. Thus far, these sequential rules could not be modelled in the simultaneous equation environment required for exact solution methods. An SA metaheuristic is developed to solve the practical scheduling model. A novel SA approach is applied where, instead of the actual solution being used for neighbourhood solution representation, the neighbours are indirectly represented by the rules used to generate neighbourhood solutions. It is also found that the initial temperature should not be a fixed value, but should be a multiple of the objective function value of the initial solution. An inverse arctan-based cooling schedule function outperforms traditional cooling schedules as it provides the required diversification and intensification behaviour of the SA. The scheduling model solves within 45 seconds and provides good, practically executable results. The metaheuristic approach to scheduling is therefore successful as the plant complexities and intricate operational philosophies can be accurately modelled using the sequential nature of programming languages and provides good approximate optimal solutions in a short solution time. Tests done with live CHF data indicate that the metaheuristic solution outperforms the current scheduling methodologies applied in the business. The implementation of the scheduler will lead to a more stable factory feed, which will increase production yields and therefore increase company profits. By reducing the amount of coal re-handling (in terms of throw-outs and load-backs at mine bunkers), the scheduler will reduce the coal handling facility’s annual operating cost by approximately R4.6 million (ZAR). Furthermore, the approaches discussed in this document can be applied to any continuous product scheduling environment. Additional information available on a CD stored at Level 3 of the Merensky Library.