A new method for monitoring flotation performance using a performance-guided autoencoder with the self-attention mechanism

Performance indicators are the core for reflecting the quality of flotation products. Real-time monitoring of these indicators is of great significance to enterprises. It remains challenging to capture the multimodal dynamic characteristics of time series data of the flotation process and effectively apply them to the monitoring task. To address this issue, this paper proposes a flotation performance monitoring method based on a performance-guided autoencoder that merges the self-attention mechanism. Firstly, the autoencoder takes the long short-term memory and the one-dimensional convolutional layer as its internal structure to parallel extract the long-term and local features of the time series data. Then, the self-attention mechanism is utilized to dynamically allocate weights to the fused features. The performance-guided autoencoder is based on unsupervised and supervised learning. The prediction error is incorporated into the loss function of the autoencoder to enhance the extracted features, making them more relevant to the monitoring task. Finally, the features extracted by the autoencoder are sent to the predictor module for real-time monitoring of performance indicators. The MAE, RMSE, and R² of the proposed method on the zinc flotation test data are 0.2475, 0.3433, and 0.7643, respectively, outperforming other existing advanced techniques. The experimental results verify the effectiveness and superiority of this method.