No abstract

Phosphate removal is a crucial objective in wastewater engineering to reduce harmful environmental impacts like eutrophication. Adsorption, a low-cost and efficient process for phosphate abatement, primarily relies on trapping phosphate on lowsolubility solid surfaces. Metal-based materials, due to their abundance, low cost, environmental friendliness, and chemical stability, are considered the most promising phosphate adsorbents. However, the synthesis of appropriate adsorbents is complex and timeconsuming. In addition, the diverse textural properties, the presence of various metals, and the selection of adsorption parameters make it challenging to the underlying mechanism of phosphate adsorption. In this study, we compiled a data set including 1800 data points mined from 128 peer-reviewed papers and adopted machine learning (ML) to systematically evaluate phosphate adsorption concerning textural properties, metal compositions, and adsorption parameters. We applied three different tree-based algorithms, including random forest (RF), decision trees (DTs), and extreme gradient boosting (XGBoost), to guide the design of adsorbents and predict the phosphate adsorption performances. Among the three algorithms, RF showed the best predictive performance with a high R 2 of 0.984 and a low root-mean-squared error (RMSE) of 0.650. Feature importance, based on the Shapley values, demonstrated the contributions of adsorbents' textural properties (e.g., surface area), adsorption parameters, and metal types in the order of precedence of phosphate adsorption, providing critical insights into guiding adsorbents design and synthesis for phosphate adsorption applications.

Recommendation algorithms save a lot of valuable time for people to get the information they are interested in. However, the feature calculation and extraction process of each machine learning or deep learning recommendation algorithm are different, so how to obtain various features with different dimensions, i.e., how to integrate the advantages of each model and improve the model inference efficiency, becomes the focus of this paper. In this paper, a better deep learning model is obtained by integrating several cutting-edge deep learning models. Meanwhile, to make the integrated learning model converge better and faster, the parameters of the integrated module are initialized, constraints are imposed, and a new activation function is designed for better integration of the sub-models. Finally, the integrated large model is distilled for knowledge distillation, which greatly reduces the number of model parameters and improves the model inference efficiency.

No abstract