Machine learning for high-precision simulation of dissolved organic matter in sewer: Overcoming data restrictions with generative adversarial networks.

Understanding the transformation process of dissolved organic matter (DOM) in the sewer is imperative for comprehending material circulation and energy flow within the sewer. The machine learning (ML) model provides a feasible way to comprehend and simulate the DOM transformation process in the sewer. In contrast, the model accuracy is limited by data restriction. In this study, a novel framework by integrating generative adversarial network algorithm-machine learning models (GAN-ML) was established to overcome the drawbacks caused by the data restriction in the simulation of the DOM transformation process, and humification index (HIX) was selected as the output variable to evaluate the model performance. Results indicate that the GAN algorithm's virtual dataset could generally enhance the simulation performance of regression models, deep learning models, and ensemble models for the DOM transformation process. The highest prediction accuracy on HIX (R2 of 0.5389 and RMSE of 0.0273) was achieved by the adaptive boosting model which belongs to ensemble models trained by the virtual dataset of 1000 samples. Interpretability analysis revealed that dissolved oxygen (DO) and pH emerge as critical factors warranting attention for the future development of management strategies to regulate the DOM transformation process in sewers. The integrated framework proposed a potential approach for the comprehensive understanding and high-precision simulation of the DOM transformation process, paving the way for advancing sewer management strategy under data restriction.

Preparation and function of composite asymmetric chitosan/CM-chitosan membrane.

A novel composite asymmetric chitosan/CM-chitosan membrane (C-P-C) was prepared, the top-layer was chitosan (CS), the intermediate was PVA, and the substrate was carboxymethyl chitosan (CM-CS). C-P-C membrane had capability in mechanical strength, light transparence, vapor permeability, and wound skin joining. The CS and CM-CS in C-P-C membrane were selected by series independent experiments, respectively. CS (MW 90,000 Da) had the highest antibacterial activity for E.coli. CM-CS had biocompatibility, no cytotoxicity, and had the activity of promoting growth of human skin fibroblast and inhibiting the growth of keloid fibroblast. The normal skin fibroblast can growth on the CM-CS surface of C-P-C, and have no conglomeration in higher cell density, and the keloid fibroblast could not growth on CM-CS surface of C-P-C. The animal experiment demonstrated that wound, covered with the C-P-C membrane, was hemostatic, healing quickly and had histocompatibility. The results indicated that the C-P-C membrane could be used as dressing of skin repair, and had the potential in promoting wound healing and inhibiting the keloid formation.