The production of hydrogen is one way to meet the ever-increasing demand for cleaner and renewable energy sources. In this study, various machine learning techniques such as Levenberg–Marquardt neural networks (LMNN), scaled conjugate gradient descent neural networks (SCDNN), and support vector machine (SVM) in comparison with nonlinear regression model (NLM) and response surface model (RSM) were employed for modeling photocatalytic hydrogen production from ethanol using copper oxide (CuO) nanoparticles as photocatalyst. The effects of input parameters such as the irradiation time, the CuO content, the catalyst dosage, and the ethanol concentration on hydrogen production were considered in the modeling process. Optimized network configurations of 4-12-1 and 4-5-1 representing the input nodes, hidden neurons, and output node were used for the LMNN and SCDNN, respectively. Both the LMNN and SCDNN show superior prediction of hydrogen production compared with the SCDNN, SVM, NLM, and RSM as indicated by the high R values of 0.998 and 0.997 for LMNN and SCDNN, respectively. The LMNN displayed the best prediction of hydrogen production with R value of 0.998. The sensitivity analysis shows that all the input parameters influenced the LMNN model output. However, the predicted hydrogen from the LMNN model was best influenced by the irradiation time.

氢的生产是满足对清洁和可再生能源不断增长的需求的一种方法。在这项研究中，与非线性回归模型（NLM）和响应面模型（ （RSM）用于模拟使用氧化铜（CuO）纳米粒子作为光催化剂从乙醇产生光催化制氢的过程。在建模过程中考虑了输入参数（例如辐照时间，CuO含量，催化剂用量和乙醇浓度）对产氢的影响。4-12-1和4-5-1的最佳网络配置代表输入节点，隐藏的神经元，和输出节点分别用于LMNN和SCDNN。LMNN和SCDNN都显示出比SCDNN，SVM，NLM和RSM更好的产氢预测，如高LMNN和SCDNN的R值分别为0.998和0.997。LMNN显示出最佳的产氢预测，R值为0.998。灵敏度分析表明，所有输入参数都影响了LMNN模型的输出。但是，从LMNN模型预测的氢受照射时间的影响最大。