蔗渣
产量(工程)
贝叶斯概率
酶水解
人工智能
机器学习
贝叶斯优化
计算机科学
化学
支持向量机
数学
质量(理念)
制浆造纸工业
水解
贝叶斯推理
模式识别(心理学)
超滤(肾)
生物技术
农学
作者
Yi Zhao,Wanting Su,Xuesong Tan
出处
期刊:
[Institute of Electrical and Electronics Engineers]
日期:2026-05-01
卷期号:: 62-66
标识
DOI:10.1109/net-lc70284.2026.11605738
摘要
Efficient biomass conversion requires optimized pretreatment strategies, but their design is complicated by the complex, nonlinear relationships between process variables and desired outcomes. Conventional empirical modeling and default-parameter machine learning often fail to accurately map these relationships, while exhaustive parameter tuning suffers from the curse of dimensionality and high computational overhead. This study systematically compares Quadratic Polynomial Regression with two advanced machine learning models: Support Vector Regression (SVR) and Extreme Gradient Boosting (XGBoost). Crucially, to overcome the computational bottlenecks of traditional Grid Search, we propose an algorithmically efficient framework integrating Gaussian Process-based Bayesian Optimization (via gp_minimize). By constructing a probabilistic surrogate model, this approach intelligently navigates the non-convex hyperparameter space, significantly reducing computational cost and accelerating convergence. Bayesian Optimization, executing 50 iterations with 5-fold cross-validation, significantly enhanced predictive performance. Both the Optimized SVR (RMSE = 2.453, R2= 0.962) and XGBoost (RMSE = 2.600, R2= 0.957) models demonstrated highly competitive and equally viable performance on the independent test set. Both models substantially outperformed the conventional Quadratic Regression baseline (RMSE = 4.121, R2= 0.892). The proposed Bayesian Optimization framework provides a robust, high-fidelity surrogate model with superior computational efficiency, offering a powerful data-driven tool to guide rapid process parameter screening and minimize both computational and experimental workloads in diverse scientific domains.
科研通智能强力驱动
Strongly Powered by AbleSci AI