This study develops a probabilistic neural network (PNN) based on 2165 laboratory data incorporating mixture constituents and environmental conditions parameters to estimate the natural carbonation depth of concrete. The high prediction accuracy and certainty level of the model, achieving a testing R of 0.95 and negative log-likelihood of 1.82, enabled performing numerical simulations of carbonation depth and rate evolution of 576 binary and ternary blends for 12 years of exposure under different climatic conditions of two cities, Toronto and Vancouver. Mixtures containing ordinary Portland cement (OPC), fly ash, ground granulated blast furnace slag (GGBFS), and limestone calcined clay with water-to-binder ratios of 0.35–0.55, OPC replacement levels of 0–0.6, and binder contents of 350–500 kg/m were studied. Results indicated that OPC-GGBFS binary binders and OPC-GGBFS-fly ash ternary binders showed the lowest natural carbonation rates of 1.03–4.78 and 1.10–3.76 mm/year, respectively. The uncertainty quantification of model's predictions can facilitate informed and confident decision-making for designing durable and sustainable concrete mixtures.

中文翻译：

---

结合 SCM 的低碳混凝土自然碳化的概率深度学习预测


本研究开发了一种基于 2165 个实验室数据的概率神经网络 (PNN)，其中结合了混合物成分和环境条件参数，以估计混凝土的自然碳化深度。该模型的预测精度和确定性水平很高，测试 R 为 0.95，负对数似然为 1.82，能够对 576 种二元和三元混合物在不同气候条件下暴露 12 年的碳化深度和速率演变进行数值模拟多伦多和温哥华两个城市。含有普通硅酸盐水泥 (OPC)、粉煤灰、磨碎的粒化高炉矿渣 (GGBFS) 和石灰石煅烧粘土的混合物，水与粘合剂的比例为 0.35–0.55，OPC 替代水平为 0–0.6，粘合剂含量为 350研究了 –500 kg/m。结果表明，OPC-GGBFS 二元粘合剂和 OPC-GGBFS-粉煤灰三元粘合剂表现出最低的自然碳化率，分别为 1.03-4.78 和 1.10-3.76 毫米/年。模型预测的不确定性量化可以促进明智且自信的决策，以设计耐用且可持续的混凝土混合物。