基于PSO-PLS煤泥浮选加药量预测模型

A PSO-PLS Based Prediction Model for Reagent Dosage in Coal Slime Flotation

摘要: 煤泥浮选药剂添加量的精确控制对浮选效果至关重要，是智能浮选的重要因素，也是近几年浮选智能科研工作者的研究课题。药剂量添加不当会导致浮选精煤灰分存在较大波动。影响浮选药剂添加量的因素众多，本文考虑入料浓度、入料流量、补水量、精煤灰分和尾煤灰分这五种因素对煤泥浮选药剂添加量的影响，提出了一种基于粒子群优化(PSO)偏最小二乘(PLS)算法的煤泥浮选起泡剂和捕收剂加药量预测模型。通过对比PCA、PLS和PSO-PLS三种算法的预测效果，发现PSO-PLS模型在预测精度和稳健性上表现优异，均方差、均方根误差、平均绝对百分比误差显著低于前两者，捕收剂预测R²值达到0.7863，起泡剂预测R²值达到0.8320，表明其拟合效果良好。实验证明，PSO-PLS算法能够准确预测浮选药剂添加量，有助于实现选煤厂浮选加药过程的智能化，为进一步选煤厂智能化建设提供了技术支持。

Abstract: Precise control of reagent addition in coal slime flotation is crucial for achieving optimal flotation performance. It is a key factor in intelligent flotation and has been a research focus for flotation intelligence researchers in recent years. Improper reagent addition can lead to significant fluctuations in the ash content of the flotation concentrate. Many factors influence the amount of reagent added, including feed concentration, feed flow rate, added water volume, concentrate ash content, and tailings ash content. This paper proposes a predictive model for the addition of frothers and collectors in coal slime flotation based on the Particle Swarm Optimization (PSO) and Partial Least Squares (PLS) algorithms. By comparing the predictive performance of PCA, PLS, and PSO-PLS algorithms, it was found that the PSO-PLS model excels in prediction accuracy and robustness, with significantly lower mean square error, root mean square error, and mean absolute percentage error compared to the other two methods. The R² value for collector prediction reached 0.7863, and for frother prediction, it reached 0.8320, indicating good fitting performance. Experiments demonstrate that the PSO-PLS algorithm can accurately predict reagent addition, aiding in the intelligent reagent dosing process in coal preparation plants and providing technical support for further intelligent construction of coal preparation plants.