Data mining, mining data: energy consumption modelling
As increasingly more data tracking production and business processes continue to be collected, mines are facing a problem seen in other businesses: being data-rich while information-poor. As additional efforts and technology are placed at developing even more information sources, a new technological focus should emerge: how to concentrate data into information; analyze information sufficiently to become knowledge; and finally, act on that knowledge (data®information®knowledge®action). New technologies, developed in non-mining industries, have begun to redress some of these data-rich–information-poor issues, specifically data warehousing (the enabling technology) and data mining (the analytical technology). An approach to develop applications and skills, wherein data is transformed into action, continues to be performed and tested at the Mining Information Systems and Operations Management (MISOM) lab at the University of Arizona. The data-to-action approach was exercised in the development of an energy consumption model (ECM), in partnership with a major US-based copper mining company, two software companies, and the MISOM lab, funded by the U.S. Department of Energy (DOE). The project, called Infrastructure for Integrated Data Environments and Analysis(IIDEA) for Mining and Processing Systems, began as a oneyear pilot study that used a copy of the 1.2 Terabyte corporatedata warehouse containing all records from every major information system (IS) used at all the operations of the partner mining company.
The data-to-action approach begins by integrating several key data sources using data warehousing techniques, namely the highly granular fleet management system (FMS, namely Dispatch®) and all cost transactions for the past four years from the enterprise system (ES, namely Ellipse®). The project began by increasing the existing level of integration and data cleaning. The information step involved the creation of online analytical processing (OLAP) cubes to investigate the data and identify a subset of several million records. Data mining algorithms, mostly neural-network based, were applied using the information that was isolated by the OLAP cube. The data mining results showed that traditional cost drivers of energy consumption, namely tons and distance for diesel, and tons for kWhrs, are poor predictors. A comparison was made between the traditional means on predicting energy consumption and the prediction formed using data mining. Traditionally, in the mines for which data were available, monthly averages of tons and distance are used to predict diesel fuel consumption. New information technology can be used to incorporate many more variables into the budgeting process, whereby far more accurate predictions can be made. The figure shows the predicted (using NN) versus actual using neural networking and includes other variables such as distance travelled up or down, time of year, and truck class, predicted by week. The data mining results are far closer to actual than when using traditional means of prediction. The final step in evolving data into action is using the newly created knowledge. The most valuable knowledge will not generate real value unless it is acted upon. Other business sectors were transformed by IT only once workflows were re-engineered to take advantage of the new capabilities. The project undertook a deliberate work and data-flow mapping of the budgeting process at the mines under study.An idealized workflow was then engineered considering data availability, technical skills of the local personnel, and cultural considerations. The ECM was developed to help mine planners improve the prediction of energy use in the materials handling system.
Business management processes, such as improvement initiatives and mine engineering, can be greatly improved through more data integration, measure development, and workflow analysis. This has been the experience in other industries such as in retail and marketing. The enabling technology is now available. What remains is solidifying and deploying data to action procedures.