Prediction of microalgae harvesting efficiency and identification of important parameters for ballasted flotation using an optimized machine learning model

Abstract

Ballasted flotation is an innovative and effective technique for the separation and recovery of microalgae. However, conventional experimental approaches to determine the optimal harvesting efficiency of microalgae are often inefficient and subjective, largely due to the varying properties of microalgae, types of ballasted agents (low-density materials, LDMs), and operational conditions. This study aims to develop a machine learning approach to establish the relationship between various features and harvesting efficiency in ballasted flotation, offering new insights for achieving efficient microalgal harvesting. The results showed that the performance of the Backpropagation Neural Network (BPNN) model outperformed other machine learning models examined in the study. To further enhance the predictive accuracy of the BPNN model, two additional optimization algorithms, Genetic Algorithm (GA) and Particle Swarm Optimization (PSO), were used to optimize the initial parameters of BPNN model. The findings demonstrated that both optimization models effectively improved the predictive ability of the BPNN model, with GA-BPNN exhibiting smaller testing Mean Absolute Error and Root Mean Square Error values (0.041 and 0.007, respectively), and a higher testing R² value (0.923), indicating superior performance compared to PSO-BPNN. SHAP analysis identified that microalgal concentration and the diameter of LDMs were the two most influential parameters affecting microalgal harvesting. Finally, experimental validation of microalgae harvesting confirmed the model's accuracy, with results falling within a 5 % error margin of the predicted values. These insights obtained through machine learning analysis can facilitate the development of high-throughput experimental designs, which can significantly enhance the harvesting efficiency of microalgae.