Feed-Forward Neural Network Soft-Sensor Modeling of Flotation Process Based on Particle Swarm Optimization and...

Area of Science:

• Mineral Processing
• Artificial Intelligence
• Chemical Engineering

Background:

• Flotation processes require accurate prediction of key indicators like concentrate grade and tailings recovery rate.
• Existing soft-sensor models face challenges with convergence speed and local optima.
• Feed-forward neural networks (FNNs) are utilized for soft-sensing but require effective parameter optimization.

Purpose of the Study:

• To propose an optimized feed-forward neural network (FNN) based soft-sensor model for predicting flotation process indicators.
• To enhance the optimization capability and prediction accuracy of soft-sensor models using a hybrid algorithm.
• To address the limitations of the gravitational search algorithm (GSA) by integrating the particle swarm optimization (PSO) algorithm.

Main Methods:

• A hybrid optimization algorithm combining PSO and GSA was developed to improve GSA's convergence velocity and avoid local optima.
• The hybrid algorithm was used to optimize the parameters of the FNN soft-sensor model.
• Simulation studies were conducted to evaluate the performance of the proposed model.

Main Results:

• The hybrid PSO-GSA algorithm demonstrated improved convergence speed and prediction accuracy compared to standalone algorithms.
• The optimized FNN soft-sensor model achieved superior generalization and prediction accuracy for concentrate grade and tailings recovery rate.
• The proposed model meets the requirements for online soft-sensing in real-time flotation process control.

Conclusions:

• The hybrid PSO-GSA optimized FNN soft-sensor model is effective for real-time prediction of flotation process indicators.
• This approach offers a robust solution for improving the efficiency and control of flotation operations.
• The study highlights the potential of hybrid metaheuristic algorithms in enhancing soft-sensor applications in mineral processing.