Time-Series Prediction of Silicon Content in Froth Flotation Based on MI-PCA and Improved LSTM

With the continuous advancement of industrial automation and intelligence, the accurate prediction of silicon content in the froth flotation process has become increasingly important. This paper proposes a time series forecasting method based on Mutual Information and Principal Component Analysis (MI-PCA) and an improved Long Short-Term Memory (LSTM) network, aiming to enhance the accuracy and efficiency of silicon content prediction in froth flotation. Firstly, the feature selection of raw data is performed by MI-PCA method, which effectively reduces the data dimensions and extractes the most influential features on the silicon content changes. Then, the improved LSTM model is applied to the filtered features to construct a model capable of handling time series data and predicting future silicon content changes. This model is able to utilize historical data from the past 15 hours to predict the silicon content in the next 3 hours, demonstrating adaptability to the complex dynamic changes in the flotation process. Comparative experimental results with traditional methods such as LSTM, Recurrent Neural network (RNN) and eXtreme Gradient Boosting (XGBoost) show that the proposed method outperforms existing methods in evaluation metrics such as Root Mean Square Error (RMSE) and Mean Absolute Error (MAE), validating the effectiveness and superiority of the model. This study not only provides technical support for the optimization of froth flotation process but also offers a new solution for time series forecasting problems in other industrial processes.