A comparative study of prediction methods for semi-autogenous grinding mill throughput

Created by W.Langdon from gp-bibliography.bib Revision:1.7675

@Article{GHASEMI:2024:mineng,

• author = "Zahra Ghasemi and Frank Neumann and Max Zanin and John Karageorgos and Lei Chen",
• title = "A comparative study of prediction methods for semi-autogenous grinding mill throughput",
• journal = "Minerals Engineering",
• volume = "205",
• pages = "108458",
• year = "2024",
• ISSN = "0892-6875",
• DOI = "doi:10.1016/j.mineng.2023.108458",
• URL = "https://www.sciencedirect.com/science/article/pii/S0892687523004727",
• keywords = "genetic algorithms, genetic programming, Prediction, Grinding mill, Throughput, Machine learning",
• abstract = "The mining industry is experiencing a growing amount of stored production data, yet the full potential of these datasets in process modelling remains unexplored. Semi-autogenous grinding (SAG) mills are extensively used in the grinding circuit of mining plants. Precise prediction of SAG mill throughput can result in significant economic benefits, as it can be used for better parameter settings to achieve higher throughputs. Furthermore, the development of an accurate throughput prediction model can assist in informed decision-making for long-term planning. The model's ability to reveal the overall effect of various inputs and estimate the potential throughput change associated with altering each input, can be helpful for determining whether to invest in altering inputs that are laborious and expensive. Numerous SAG mill models have been investigated in the literature; however, a few studies were aimed at forecasting mill throughput. In this research the most accurate prediction model for SAG mill throughput will be investigated through comparing six machine learning models, including genetic programming, recurrent neural networks, support vector regression, regression trees, random forest regression, and linear regression. To achieve this purpose, a real-world data set comprised of 20,161 records from a gold mining complex in Western Australia is investigated and the effective parameters are identified as SAG mill turning speed, power draw of SAG mill, inlet water, and input particle size. As the data set is in the form of time series, the time-dependent nature of the data is considered for prepossessing, model selection, and final comparison. Specially for the first time in this research, delays in data are investigated and used to improve prediction performance. Moreover, hyperparameter tuning is performed to determine the best parameter setting for each model prior to implementation. The comparison results demonstrate that the recurrent neural network is the most accurate prediction model, followed by genetic programming and support vector regression. The genetic programming approach is also able to provide a mathematical equation for the SAG mill throughput prediction, which is highly valued by experts in the industry. Sensitivity analysis revealed that the two factors that most significantly affect SAG mill throughput are turning speed and inlet water. It is anticipated that the SAG mill throughput will rise as the SAG mill turning speed increases and the input water decreases",

}

Genetic Programming entries for Zahra Ghasemi Frank Neumann Max Zanin John Karageorgos Lei Chen

Citations