Gold ore deposits are decreasing in grade and are becoming more complex in nature. The Knelson gravity concentrator is an attractive option for processing these ores, though it is normally used for free-milling, because they are already widely accepted in the industry and are relatively easy to operate. However, more work needs to be done to optimise the operating conditions for these ore types. In this investigation data from different sources are pooled an a apley dditive exlanations (SHAP) analysis performed to quantify the effects of four individual variables on the performance of the concentrator. These variables were bowl speed, fluidisation water rate, solids feed rate and feed density. Using these variables as predictors, random forest models could explain 85% of the variance of the grade and 74% of the variance of the recovery of the material. The influence of each predictor could be quantified in terms of its percentage contribution to the Shapley values of the models (R-values). For the feed SG, bowl speed, solids feed and fluidisation water flow rate, these were 56.2%, 16.4%, 15.2% and 12.2% for the grade and 39.0%, 26.9%, 17.1% and 16.9% for recovery, respectively. Similar complementary results were obtained with a tree SHAP analysis of the random forest models. Partial dependence plots revealed that the relationships between the predictors and the grade and recoveries were generally strongly nonlinear.

中文翻译：

---

操作变量对 Knelson 选矿厂回收率和品位影响的 Shapley 分析

金矿床的品位正在下降，性质也变得更加复杂。尼尔森重力选矿机是处理这些矿石的一个有吸引力的选择，尽管它通常用于自由研磨，因为它们已经在行业中被广泛接受并且相对易于操作。然而，需要做更多的工作来优化这些矿石类型的操作条件。在这项调查中，来自不同来源的数据被汇集起来，并进行 apley 加法解释 (SHAP) 分析，以量化四个单独变量对集中器性能的影响。这些变量是转鼓速度、流化水速率、固体进料速率和进料密度。使用这些变量作为预测变量，随机森林模型可以解释 85% 的品位方差和 74% 的材料回收率方差。每个预测变量的影响可以根据其对模型 Shapley 值（R 值）的贡献百分比来量化。对于进料 SG、转鼓速度、固体进料和流化水流量，品位分别为 56.2%、16.4%、15.2% 和 12.2%，回收率分别为 39.0%、26.9%、17.1% 和 16.9%。通过随机森林模型的树 SHAP 分析获得了类似的互补结果。偏相关图显示预测变量与品位和回收率之间的关系通常呈强非线性关系。