Accurate estimation of annual output energy yield ($E_{\text{out,annual}}$) for perovskite/silicon (PSK/cSi) tandem solar cells is pivotal in assessing their suitability for building-integrated photovoltaics (BIPV). This study pioneers five machine learning models of ensembles of trees, Gaussian process regressions, regression trees, support vector machines, and artificial neural networks (ANN) to predict output power density and compute $E_{\text{out,annual}}$ for 2T, 3T, and 4T PSK/cSi tandem configurations in Japan's outdoor conditions. Seven predictive inputs of visible-light solar irradiance, near-infrared-light solar irradiance, incident solar spectrum angle, solar module temperature, perovskite thickness, perovskite bandgap, and terminal of tandem configuration (T) drive the machine learning models. These models optimize $E_{\text{out,1-month}}$ predictions using k-fold cross-validation and Bayesian algorithms, showcasing superior precision in $E_{\text{out,annual}}$ prediction compared to prior models. The ANN model emerges as the best model, displaying the minimal error in predicting $E_{\text{out,1}\hspace{0.17em}\text{month}}$, used to estimate $E_{\text{out,annual}}$ across five Japanese locations (Gifu, Naganuma, Okinoerabu, Tosu, and Tsukuba). Results from these locations in blue-rich solar spectrum zones identify the 4T PSK/cSi tandem configuration, featuring the most outstanding mean maximal $E_{\text{out,annual}}$ (93.63, 263.02, 153.59, and 91.75 kWh m⁻² for the east, rooftop, south, and west directions), as the prime candidate for BIPV applications.

中文翻译：

---

基于机器学习预测的年度室外发电量，分析和选择建筑一体化光伏发电的最佳钙钛矿/硅串联配置

准确估算年发电量（${乙}_{\text{出，每年}}$）对于钙钛矿/硅（PSK/cSi）串联太阳能电池来说，对于评估其对建筑一体化光伏发电（BIPV）的适用性至关重要。这项研究开创了树集成、高斯过程回归、回归树、支持向量机和人工神经网络 (ANN) 的五种机器学习模型，用于预测输出功率密度和计算${乙}_{\text{出，每年}}$适用于日本室外条件下的 2T、3T 和 4T PSK/cSi 串联配置。可见光太阳辐照度、近红外光太阳辐照度、入射太阳光谱角、太阳能模块温度、钙钛矿厚度、钙钛矿带隙和串联配置终端 ( T ) 的七个预测输入驱动机器学习模型。这些模型优化${乙}_{\text{出去，1个月}}$使用 k 倍交叉验证和贝叶斯算法进行预测，展示了卓越的精度${乙}_{\text{出，每年}}$与之前的模型进行比较的预测。 ANN 模型成为最佳模型，预测误差最小${乙}_{\text{输出,1}\hspace{0.17em}\text{月}}$，用于估计${乙}_{\text{出，每年}}$跨越五个日本地点（岐阜、长沼、冲永良部、鸟栖和筑波）。来自富含蓝光的太阳光谱区域中这些位置的结果确定了 4T PSK/cSi 串联配置，具有最突出的平均最大${乙}_{\text{出，每年}}$（东、屋顶、南、西方向分别为93.63、263.02、153.59 和 91.75 kWh m ^{−2 ），作为 BIPV 应用的主要候选者。}