Scholarlink

<p>Certain green chemistry processes rely on the solubility of compounds in supercritical CO<sub>2</sub>. This work constructs and evaluates several prediction models based on solubility, which include temperature, pressure, and density as input features. XGBoost hyperparameter optimization with Gaussian process priors is used along with a deep learning model optimized by combining grid and random search with cross-validation. In addition, we propose a novel hybrid model based on the integration of Grey Wolf Optimizer and Differential Evolution, referred to as GWO-DE-LSTM. Comprehensive feature importance analyses highlight different representation learning outcomes: XGBoost, with an R2 of 0.881, uses features in a balanced manner; DL attributed 72.6% importance to pressure and achieved R2 = 0.895; while GWO-DE-LSTM, surpassing others at R2 0.931, uniquely emphasizes temperature with 57.1% importance. These findings support why this hybrid model performed better than traditional approaches. The results suggest that these evolutionary strategies do enhance accuracy regarding estimating the solubility of CO<sub>2</sub> in supercritical conditions but also provide different interpretations regarding feature relationships.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A novel GWO-DE-LSTM hybrid model for predicting statin drug solubility in supercritical carbon dioxide: a comparative analysis with traditional machine learning approaches

  • 期刊论文

  • 发表日期：

    2025年6月28日

查看全文
  • Ali Damansabz,
  • Mostafa Khajeh,
  • Jamshid Piri,
  • Mansour Ghaffari-Moghaddam

摘要

Certain green chemistry processes rely on the solubility of compounds in supercritical CO2. This work constructs and evaluates several prediction models based on solubility, which include temperature, pressure, and density as input features. XGBoost hyperparameter optimization with Gaussian process priors is used along with a deep learning model optimized by combining grid and random search with cross-validation. In addition, we propose a novel hybrid model based on the integration of Grey Wolf Optimizer and Differential Evolution, referred to as GWO-DE-LSTM. Comprehensive feature importance analyses highlight different representation learning outcomes: XGBoost, with an R2 of 0.881, uses features in a balanced manner; DL attributed 72.6% importance to pressure and achieved R2 = 0.895; while GWO-DE-LSTM, surpassing others at R2 0.931, uniquely emphasizes temperature with 57.1% importance. These findings support why this hybrid model performed better than traditional approaches. The results suggest that these evolutionary strategies do enhance accuracy regarding estimating the solubility of CO2 in supercritical conditions but also provide different interpretations regarding feature relationships.