A Flotation Process Performance Perception Method Based on TPA-BiLSTM and a Working Condition Knowledge-Guided Attention Mechanism

The flotation process is a key link in the mineral processing field, and accurately perceiving its performance is of great significance. The pulp grade of the key cell is an important performance indicator of the flotation process. This paper proposes a method for perceiving flotation process performance based on the Temporal Pattern Attention mechanism and the Bidirectional Long Short-Term Memory network (TPA-BiLSTM), which additionally incorporates a working condition knowledge-guided attention mechanism. In this method, the BiLSTM captures temporal sequence dependencies, and the TPA precisely selects key information, capturing complex dynamic patterns across time steps. The working condition knowledge-guided attention mechanism module first retrieves similar patterns to the current operation condition through case matching. Then, it extracts working condition knowledge from these retrieved similar patterns. On this basis, a weight allocation mechanism is constructed to adjust the contribution of each dimension feature in the feature representation vector. The experimental results show that the RMSE and R2 of the proposed model on the zinc foam flotation data test set are 0.2826 and 0.8464. This result verifies the reliability of the proposed model in perceiving flotation performance.