| Small-Data Gaussian Process Modeling and Genetic-Algorithm Optimization of Bi2WO6-x Photocatalysis for Cefixime Removal from Water |
| کد مقاله : 1308-IWWA |
| نویسندگان |
|
آریا گردان شکن1، شایان عبائی2، شکیبا عربیان3، رسول گردان شکن4، امید توکلی *5 1دانشجوی دکتری مهندسی شیمی دانشگاه تهران مدیرعامل شرکت بین المللی انبوهش آریای اصفهان 2دانشجوی کارشناسی ارشد مهندسی شیمی دانشگاه تهران 3دانشجوی دکتری مهندسی شیمی دانشگاه صنعتی اصفهان مدیر واحد تحقیق و توسعه شرکت بین المللی انبوهش آریای اصفهان 4مدیر عامل شرکت مهندسی تحقیقات بسپار بهینه اصفهان رئیس هیئت مدیره شرکت بین المللی انبوهش آریای اصفهان 5عضو هیئت علمی گروه مهندسی شیمی دانشگاه تهران |
| چکیده مقاله |
| Cefixime and other antibiotic residues can persist through conventional wastewater treatment, motivating the development of advanced removal processes and data-efficient optimization strategies. In this study, Bi2WO6-x was synthesized through a solvothermal route and evaluated for photocatalytic cefixime removal. A 57-run experimental dataset combined 27 full-factorial conditions with 30 Latin hypercube samples spanning initial cefixime concentration (10-20 mg/L), catalyst dosage (0.10-0.30 g/L), and light intensity (200-600 W/m2). The response was cefixime removed per gram of catalyst after a fixed 180 min reaction. Gaussian process regression was optimized with Optuna and evaluated using nested 5×5-fold cross-validation. The out-of-fold predictions achieved R2 = 0.9842, RMSE = 2.5069 mg/g, MAE = 1.3989 mg/g, and 96.49% coverage of the nominal 95% predictive interval. Two-dimensional partial dependence plots showed that the mass-normalized response increased toward higher initial concentration and light intensity and lower catalyst dosage, while final-model uncertainty maps showed only modest variation within the experimental domain. A DEAP genetic algorithm converged in all eight runs to 20 mg/L, 0.10 g/L, and 600 W/m2, with a final-model prediction of 117.3169 ± 0.1439 mg/g; the corresponding measured training-set value was 117.3403 mg/g. X-ray diffraction and scanning electron microscopy provided complementary structural and morphological characterization. The results demonstrate a compact workflow combining experiment, probabilistic machine learning, uncertainty analysis, and bounded optimization for photocatalytic water-treatment studies. |
| کلیدواژه ها |
| Cefixime, Bi2WO6-x, photocatalysis, Gaussian process regression, uncertainty, genetic algorithm |
| وضعیت: پذیرفته شده برای ارسال فایل های ارائه پوستر |
