随着国家政策和新一代人工智能技术的持续牵引,矿山智能化研究取得突破,其中,选煤厂智能化建设受到高度关注,煤泥浮选智能控制技术一直是阻碍选煤厂智能化建设的关键瓶颈之一。以煤泥浮选数据生命周期为主线,从浮选精煤/尾煤灰分在线预测、浮选药剂智能添加和煤泥浮选系统智能决策3个角度综述了煤泥浮选智能控制技术的研究进展,并展望未来煤泥浮选智能控制技术发展趋势。浮选精煤灰分在线预测困难重重,单一视觉特征信息并不可靠,尾矿灰分的预测技术相对更加成熟可靠;浮选药剂添加量受多个过程变量同时制约,模型性能在整个工况区间的自适应性和泛化能力还需进一步提升;当前浮选工业系统智能控制技术的进一步发展严重受限于浮选精煤/尾煤灰分等指标的预测精度、传感器检测精度、药剂添加精度等因素。浮选过程数据集维度更高,难以建立可靠的知识库,以深度学习为代表的新一代人工智能技术能适应这类数据结构。此外,已有浮选监测系统只针对特定矿物,唯一性较高。未来浮选智能控制系统应集中攻克指标预测、传感器检测精度等方面限制,建立多煤种、模板化的煤泥浮选智能控制资料大数据集和大模型。

With the continuous traction of Chinese government policies and the new artificial intelligence technology, the research of mineintelligence has continued to make breakthroughs in recent years. The intelligent construction of coal preparation plant as a part of intelligent mine has received great attention, among which, the intelligent control technology of coal flotation has been one of the key bottleneckshindering the intelligent construction of coal preparation plant. In this paper, the life cycle of coal slime flotation data was taken asthe main research line, the research progress of coal flotation intelligent control technology was reviewed from three perspectives: onlineprediction of coal flotation concentrate/ tailings ash content, intelligent addition of the flotation regents and intelligent decision-makingof coal flotation system, and the research tendency of coal flotation intelligent control was looked forward to the future. The online prediction of concentrate ash content is still difficult, and the single computer visual feature information of froth image is not reliable, the prediction technology of tailings ash content is relatively more reliable. The addition of flotation regents is limited by multiple flotation conditionvariables at the same time, and the adaptability and generalization ability of model performance in the entire working condition intervalneed to be further improved. The current research on flotation intelligent control technology is limited by the prediction accuracy of coal flotation concentrate/ tailings ash content, sensor detection accuracy, and agent addition accuracy. The flotation process dataset is more dimensional, making it difficult to establish a reliable knowledge base. The new generation of artificial intelligence technology represented bydeep learning can adapt to this kind of data structure. In addition, the existing flotation monitoring system only targets specific minerals,with high uniqueness. In the future, the coal flotation intelligent control system should focus on overcoming the limitations of index prediction and sensor detection accuracy, and establish a large dataset and large model of multi-coal and templated intelligent control data.

0 引言

1 浮选数据生命周期与智能控制技术概述

2 煤泥浮选系统智能控制

   2.1 浮选精煤/尾煤灰分在线预测

   2.2 浮选药剂智能添加

   2.3 煤泥浮选过程智能决策

3 结语及展望