Gholamzadeh Erfan, Ghaemi Ahad, Shokri Abolfazl, Heydari Bahman
School of Chemical, Petroleum and Gas Engineering, Iran University of Science and Technology, Tehran, Iran.
Faculty of Mechanical Engineering, University of Tehran, Iran.
Heliyon. 2024 Dec 24;11(1):e41450. doi: 10.1016/j.heliyon.2024.e41450. eCollection 2025 Jan 15.
In order to lower total energy consumption, this study focuses on optimizing energy use in refinery boilers. Using Aspen HYSYS simulations and modeling approaches like Artificial Neural Networks (ANNs) and Response Surface Methodology (RSM), data from 579 days of boiler operation was gathered and examined. Radial Basis Function (RBF) and Multi-Layer Perceptron (MLP) techniques were used in the ANN modeling. Under the same operating circumstances, Aspen HYSYS estimated an energy usage of 1,355 m³, whereas the actual consumption was 986 m³. While the R values for the ANN models were 0.98 for the RBF model and 0.99 for the MLP model, the R value derived using RSM was 0.97. Furthermore, the RBF model's performance metrics were 0.0034, whereas the MLP model's were 0.0018. The MLP model is the best choice, according to these findings. It is estimated that burning 26,000 m³ of fuel with an air supply of 23 m³/h at 25.5 °C will result in a steam flow of 525.5 tons per day at 10.5 barg and 256.5 °C. According to actual statistics, these circumstances might prevent the release of 27 tons of carbon dioxide by reducing fuel usage by over 10,000 m³ per hour. By optimizing the combustion stack's air supply, this decrease is accomplished.
为了降低总能耗，本研究着重于优化炼油厂锅炉的能源使用。利用Aspen HYSYS模拟以及人工神经网络（ANN）和响应面方法（RSM）等建模方法，收集并检查了来自579天锅炉运行的数据。在ANN建模中使用了径向基函数（RBF）和多层感知器（MLP）技术。在相同的运行条件下，Aspen HYSYS估计的能源使用量为1355立方米，而实际消耗量为986立方米。ANN模型中，RBF模型的R值为0.98，MLP模型的R值为0.99，而使用RSM得出的R值为0.97。此外，RBF模型的性能指标为0.0034，而MLP模型的性能指标为0.0018。根据这些结果，MLP模型是最佳选择。据估计，在25.5℃下，每小时供应23立方米空气燃烧26000立方米燃料，将在10.5巴表压和256.5℃下产生每天525.5吨的蒸汽流量。根据实际统计数据，通过每小时减少超过10000立方米的燃料使用量，这些情况可防止排放27吨二氧化碳。通过优化燃烧烟囱的空气供应来实现这种减少。
