Silica grade forecast in flotation processes: Evaluation and Statistical Diagnosis of Machine Learning Methods
Resumo
This study evaluates machine learning models for one-step-ahead (T + 1) silica grade forecasting in iron ore flotation, comparing autoregressive (AR) and autoregressive plus process variables (AR+Var) strategies on a private industrial dataset and a public Kaggle dataset. The evaluated models were Decision Tree, Random Forest, XGBoost, and LSTM, with a Naive persistence forecast used as baseline. The results show that the Naive method is a strong benchmark, especially in the industrial dataset, where several tree-based models did not outperform it. The best predictive result in the industrial scenario was obtained by the AR-LSTM (R2 = 0.75, MAE = 0.37), but residual diagnostics indicated remaining autocorrelation and bias. In the public dataset, Random Forest achieved consistent performance (R2 = 0.66 in both AR and AR+Var) and passed residual diagnostic tests, while LSTM achieved the highest R2 in AR (0.67) but failed residual adequacy tests.
Referências
Drumond, P., Kappes, D., De Moraes, C., et al. (2018). Soft sensor: Machine learning tradicional ou deep learning? In Seminário de Automação e TI, pages 231–242.
Fan, G., Wang, L., Cao, Y., et al. (2020). Collecting agent–mineral interactions in the reverse flotation of iron ore: A brief review. Minerals, 10(8).
Filippov, L., Severov, V., and Filippova, I. (2014). An overview of the beneficiation of iron ores via reverse cationic flotation. International Journal of Mineral Processing, 127:62–69.
Guo, X., Gao, Y., Li, Y., et al. (2021). Short-term household load forecasting based on long- and short-term time-series network. Energy Reports, 7:58–64.
Hewamalage, H., Ackermann, K., and Bergmeir, C. (2023). Forecast evaluation for data scientists: common pitfalls and best practices. Data Mining and Knowledge Discovery, 37(2):788–832.
Hyndman, R. J. and Athanasopoulos, G. (2018). Forecasting: Principles and Practice. OTexts.
McCoy, J. and Auret, L. (2019). Machine learning applications in minerals processing: A review. Minerals Engineering, 132:95–109.
Nakhaei, F., Rahimi, S., and Fathi, M. (2022). Prediction of sulfur removal from iron concentrate using column flotation froth features: Comparison of k-means clustering, regression, backpropagation neural network, and convolutional neural network. Minerals, 12(11):1434.
Oliveira, E. M. (2017). Quality prediction in a mining process. [Online].
Perera, Y. S., Ratnaweera, D., Dasanayaka, C. H., et al. (2023). The role of artificial intelligence-driven soft sensors in advanced sustainable process industries: A critical review. Engineering Applications of Artificial Intelligence, 121:105988.
Pu, Y., Szmigiel, A., and Apel, D. B. (2021). Purities prediction in a manufacturing froth flotation plant: the deep learning techniques. Neural Computing and Applications, 33:12217–12227.
Pu, Y., Szmigiel, A., Chen, J., and Apel, D. B. (2020). Flotationnet: A hierarchical deep learning network for froth flotation recovery prediction. Powder Technology, 375:317–326.
Pural, Y. E. (2023). Developing a data-driven soft sensor to predict silicate impurity in iron ore flotation concentrate. Physicochemical Problems of Mineral Processing, 59(5).
Suhasini, R., Mallick, A. K., Vasumathi, N., et al. (2015). Evaluation of flotation collectors in developing zero waste technology for processing iron ore tailings. International Journal of Engineering Research, 4(11):604–608.
Szmigiel, A., Apel, D. B., Skrzypkowski, K., et al. (2024). Advancements in machine learning for optimal performance in flotation processes: A review. Minerals, 14(4):57.
Warne, K., Prasad, G., Rezvani, S., et al. (2004). Statistical and computational intelligence techniques for inferential model development: a comparative evaluation and a novel proposition for fusion. Engineering Applications of Artificial Intelligence, 17(8):871–885.
