This study examined artificial neural networks' (ANNs) applicability in modeling optimum insulation thickness (OIT), annual total net savings (ATS), and reduction of carbon dioxide emissions (RCO₂) that result from insulating buildings. Data from insulation markets, economic parameters, fuel prices, and heating degree days (HDDs) were introduced into the model as input variables. To complete the most thorough analysis, three learning algorithms, Levenberg Marquardt (LM), Bayesian Regularization (BR), and Scaled Conjugate Gradient (SCG) were employed. Five statistical indexes were utilized to evaluate models' performances: determination coefficient (R²), root mean square error (RMSE), standard error of prediction (SEP), RMSE observations' standard deviation ratio (RSR), and average absolute percent relative error (AAPRE). Moreover, visualization techniques were used to assess the similarity between the OIT, ATS, and RCO₂ values calculated and predicted. The results obtained clearly show that the LM model outperformed the BR and SCG models in all estimations. Thereafter, the developed ANNs model was validated for different cities. Overall, this model will provide an effective and straightforward guide for people working in the field to improve thermal insulation design, analysis, and implementation worldwide.

中文翻译：

---

使用人工神经网络来估计最佳的绝缘厚度，节能和二氧化碳排放

这项研究检查了人工神经网络（ANN）在建模最佳隔热厚度（OIT），年度总净节约（ATS）和减少隔热建筑物导致的二氧化碳排放量（RCO ₂）方面的适用性。来自绝缘市场，经济参数，燃料价格和加热天数（HDD）的数据作为输入变量被引入模型中。为了完成最彻底的分析，采用了三种学习算法，即Levenberg Marquardt（LM），贝叶斯正则化（BR）和Scaled Conjugate Gradient（SCG）。利用五个统计指标评估模型的性能：确定系数（R ²），均方根误差（RMSE），预测标准误差（SEP），RMSE观测值的标准偏差比（RSR）和平均绝对百分比相对误差（AAPRE）。此外，可视化技术用于评估计算和预测的OIT，ATS和RCO ₂值之间的相似性。获得的结果清楚地表明，在所有估计中，LM模型均优于BR和SCG模型。此后，针对不同城市验证了开发的人工神经网络模型。总体而言，该模型将为在现场工作的人们提供有效而直接的指南，以改善全球的隔热设计，分析和实施。