Data-Driven Feature Decomposition Integrated Prediction Model for Dust Concentration in Open-Pit Mines

Abstract

Accurate prediction of dust in open-pit mines can serve as a foundation for implementing dust prevention and control measures. Based on the collection and monitoring of dust concentration, meteorological, and production data from open-pit mines, the changing characteristics of dust concentration and its influencing factors were analyzed. The key influencing factors of dust concentration were identified through Pearson correlation analysis. The study also systematically identified the essential and pattern characteristics of the dust time series data and utilized the variational mode decomposition (VMD) with Golden Jackal Optimization (GJO) to decompose the original dust concentration data. Combining the characteristics of dust concentration data and the concept of multimodal information integration modeling, a support vector machine (SVM)-long short-term memory (LSTM) network was chosen to build a data feature-driven dust concentration combination prediction model. The findings indicate that humidity, wind speed, stripping amount, and temperature are the primary factors influencing dust concentration. The original data on dust concentration is not only nonstationary, nonlinear, and nonperiodic but also exhibits high complexity and variability. The decomposition ensemble prediction model can accurately forecast the dust concentration in open-pit mines. Compared to SVM, LSTM, GIO-VMD-SVM, and GJO-VMD-LSTM models, the decomposition ensemble prediction model can reduce the complexity of prediction data and has a better ability to capture information. The evaluation indexes R², RMSE, and MAE of the model are 0.92559, 6.3151, and 4.5820, respectively. The prediction performance is the best.