Harmse Michael David, van Laar Jean Herman, Pelser Wiehan Adriaan, Schutte Cornelius Stephanus Lodewyk
Department of Industrial Engineering, Stellenbosch University, Stellenbosch, South Africa.
Front Artif Intell. 2022 Jul 27;5:938641. doi: 10.3389/frai.2022.938641. eCollection 2022.
The deep-level mining industry is experiencing narrowing profit margins due to increasing operating costs and decreasing production. The industry is known for its lack of dynamic control across complex integrated systems running deep underground, making IoT technologies difficult to implement. An important integrated system in a typical underground mine is the refrigeration-ventilation system. In practice, the two systems are still controlled independently, often due to a lack of continuous measurements. However, their integrated effects ultimately affect energy usage and production. This study develops and compares various machine learning prediction techniques to predict the integrated behavior of a key component operating on the boundary of the refrigeration-ventilation system, while also addressing the lack of continuous measurements. The component lacks sensors and the developed industrial machine learning models negate the effect thereof using integrated control. The predictive models are compared based on accuracy, prediction time, as well as the amount of data required to obtain the required level of accuracy. The "Support Vector Machines" method achieved the lowest average error (1.97%), but the "Artificial Neural Network" method is more robust (with a maximum percentage error of 12.90%). A potential energy saving of 215 kW or 2.9% of the ventilation and refrigeration system, equivalent to R1.33-million per annum ($82 900) is achievable using the "Support Vector Machines" method.
由于运营成本上升和产量下降，深部采矿行业的利润率正在收窄。该行业以缺乏对地下深处复杂集成系统的动态控制而闻名，这使得物联网技术难以实施。典型地下矿井中的一个重要集成系统是制冷通风系统。实际上，这两个系统仍然是独立控制的，这通常是由于缺乏连续测量。然而，它们的综合效应最终会影响能源使用和生产。本研究开发并比较了各种机器学习预测技术，以预测在制冷通风系统边界运行的关键组件的综合行为，同时也解决了连续测量不足的问题。该组件缺乏传感器，而开发的工业机器学习模型通过集成控制消除了其影响。基于准确性、预测时间以及获得所需精度水平所需的数据量对预测模型进行了比较。“支持向量机”方法实现了最低的平均误差（1.97%），但“人工神经网络”方法更稳健（最大百分比误差为12.90%）。使用“支持向量机”方法可实现215千瓦的潜在节能，或通风和制冷系统能耗的2.9%，相当于每年133万兰特（82900美元）。
