With the rise of historical data availability and increased computational power, reinforcement learning (RL) has become more prominent. It has shown the ability to outperform human knowledge in certain areas. RL is particularly valuable in control applications due to its generalization, self-adjustment, and independence from a mathematical model. Despite its potential, there are limited studies and applications focusing on using RL for mining processes. This study aims to showcase the applicability of RL in the mining industry. An algorithm was developed using proximal policy optimization (PPO) to control a simulated conventional cylindrical-conical thickener. The controller’s goal is to adjust the slurry density and solid interface height to produce a material with a high solid concentration. By regulating the thickener’s output flow and the flocculant dosage in the incoming slurry, PPO achieves this objective. Through simulation, an RL agent was trained to control the thickener efficiently, managing flocculant usage and responding appropriately to system disturbances. The study concludes that applying reinforcement learning holds promise for enhancing mining process control. However, further research is needed to fine-tune algorithm parameters, enhance design structure, optimize control, and maximize the technique’s benefits.

中文翻译：

---

通过近端策略优化控制传统连续浓缩机


随着历史数据可用性的提高和计算能力的增强，强化学习（RL）变得更加突出。它已显示出在某些领域超越人类知识的能力。强化学习由于其泛化性、自调整性以及独立于数学模型的特点，在控制应用中特别有价值。尽管有潜力，但专注于将强化学习用于挖掘过程的研究和应用仍然有限。本研究旨在展示强化学习在采矿业中的适用性。使用近端策略优化 (PPO) 开发了一种算法来控制模拟的传统圆柱形浓缩机。控制器的目标是调节浆料密度和固体界面高度，以生产具有高固体浓度的材料。通过调节浓缩机的输出流量和进入浆料中的絮凝剂剂量，PPO 实现了这一目标。通过模拟，RL 代理经过训练可以有效控制增稠剂、管理絮凝剂的使用并对系统扰动做出适当的响应。该研究的结论是，应用强化学习有望增强采矿过程控制。然而，还需要进一步研究来微调算法参数、增强设计结构、优化控制并最大限度地发挥技术的优势。