Accurate temperature prediction is crucial for various scientific and engineering applications, yet it remains challenging due to the complex relationships between spatial coordinates and temperature variations. This study addresses this challenge by exploring advanced machine learning models to improve prediction accuracy, filling the research gap in optimizing predictive models for temperature estimation in freeze drying of pharmaceuticals. We employed Support Vector Regression (SVR), Poisson Regression (POR), and Adaptive Neuro-Fuzzy Inference System (ANFIS) models, each enhanced using the Cheetah Optimizer (CO), to estimate temperature based on spatial coordinates (X, Y, Z) in the domain of process. The methodology involved training and testing these models on a dataset comprising spatial and temperature data, with a focus on improving accuracy through optimization techniques. Validation of the models showed that the CO-SVR model outperformed the others, achieving an R2 score of 0.953, Mean Squared Error (MSE) of 9.893, and Mean Absolute Error (MAE) of 2.549. This represents a significant improvement over the CO-ANFIS model, which obtained an R2 of 0.863, MSE of 23.583, and MAE of 3.912. The CO-POR model showed the lowest predictive capability, with an R2 score of 0.753, MSE of 57.608, and MAE of 6.756. These findings underscore the effectiveness of the Cheetah Optimizer in enhancing the accuracy of SVR models for temperature prediction, suggesting its potential for broader applications in predictive modeling where accuracy is paramount.

中文翻译：

---

药物冷冻干燥的计算和智能建模分析，用于预测过程中的温度


准确的温度预测对于各种科学和工程应用至关重要，但由于空间坐标和温度变化之间的复杂关系，它仍然具有挑战性。本研究通过探索先进的机器学习模型来提高预测准确性，填补了优化药品冷冻干燥温度估计预测模型的研究空白，解决了这一挑战。我们采用支持向量回归 (SVR)、泊松回归 (POR) 和自适应神经模糊推理系统 (ANFIS) 模型（每个模型均使用 Cheetah 优化器 (CO) 进行增强）来根据空间坐标（X、Y、Z）估计温度）在过程域中。该方法涉及在包含空间和温度数据的数据集上训练和测试这些模型，重点是通过优化技术提高准确性。模型验证表明，CO-SVR 模型优于其他模型，R2 得分为 0.953，均方误差 (MSE) 为 9.893，平均绝对误差 (MAE) 为 2.549。这比 CO-ANFIS 模型有显着改进，后者的 R2 为 0.863，MSE 为 23.583，MAE 为 3.912。 CO-POR 模型的预测能力最低，R2 得分为 0.753，MSE 为 57.608，MAE 为 6.756。这些发现强调了 Cheetah Optimizer 在提高 SVR 模型温度预测准确性方面的有效性，表明其在准确性至关重要的预测建模中具有更广泛应用的潜力。