Improving electricity forecasting processes on platinum mines

Abstract

Electricity is a fundamental resource used in mining, and as a result, is a large contributor to mine working cost. It is therefore important for mines to accurately budget their electricity expenditure in order to make educated financial decisions. Inaccurate electricity budgeting can have a negative effect on capital and operational planning. An important step in the electricity budgeting process is the electricity usage forecast. Electricity usage forecasting is used to predict the mine’s electricity usage based on specified system inputs, typically production. This is difficult to perform accurately due to the complexity of mining systems. Furthermore, the existing forecasting process uses overly simplified models – intensity models. These models are relatively inaccurate when compared with forecasting processes used in other industries. The negligence of detailed operational changes in the forecasting process is the main contributor to the inaccuracies observed. Therefore, a need exists for the application of technical knowledge and more advanced modelling processes to mine data. This will be done in order to develop more accurate and robust electricity forecasts with the aim of improving the mine’s financial planning and energy management. To address the problem, a methodology was developed by evaluating forecasting processes followed in other industries. This method was designed to consider: 1) model definition, 2) data collection and preparation, 3) model development and testing and 4) model interpretation. This methodology was applied to three case study mines, comprised of five deep-level, conventional mine shafts in total, with the exclusion of downstream concentrators, smelters and processing plants. Using pre-selected input variables, a variety of black-box models of different complexities were fit to the data using ordinary least squares regression for the prediction of electricity usage. The performance of these models was then compared with the use of a range of performance metrics. This was done in order to select the most accurate model for each case study. The final forecasting models were tested by using the models to forecast the electricity usage of each case study. These results were compared with the forecast produced by using the existing method in order to validate the proposed methodology. ii The application of the proposed methodology resulted in a reduction in mean absolute percentage error from 10.35% to 7.69%. This is complementary to a 37.5% reduction in root mean squared error from 0.62 GWh, using the intensity method, to 0.38 GWh using the proposed methodology. This reduction illustrates the clear improvement in forecasting accuracy when using the proposed method. Finally, an improved forecasting methodology was developed and applied to three platinum mines, comprised of five shafts in total. The new methodology involves the application and comparison of different models and offers improvements in forecasting accuracy.