This study investigates the prediction of the combustion process using an artificial neural network (ANN), and an efficient prediction methodology is introduced. In particular, the conditions for using hydrogen as an additive to a turbo-charged gasoline direct injection (T-GDI) engine are discussed. Research to predict the physical phenomena using ANNs has been actively conducted for various applications, including internal combustion engines. However, a large amount of data must be collected under various conditions to establish these predictions. Furthermore, the prediction of complex phenomena such as engine-combustion processes mandates data collection under diverse conditions. It is therefore very costly and time-consuming to obtain these data experimentally under a wide range of conditions. However, the methodology introduced in this study can enable effective prediction of complex combustion processes, such as hydrogen-added combustion, with minimal experimental data. To implement this methodology, the target engine was modeled using commercial 1D engine-simulation software GT-Power based on certain experimental results obtained under select conditions. The data for the ANN training under an expanded range of conditions were obtained using the GT-Power engine model. According to the obtained data, the ANN model for prediction of the hydrogen-added combustion processes in the T-GDI engine was constructed, and its results were compared with the experimental results. A reasonable agreement between the compared results was observed, which demonstrated the validity and reliability of the prediction model. The constructed ANN combustion model has the potential that it can be applied to transient conditions or used as a virtual sensor, unlike general combustion models, and this study presented an economical and efficient way to build such a model.

这项研究调查了使用人工神经网络（ANN）对燃烧过程的预测，并介绍了一种有效的预测方法。特别地，讨论了使用氢气作为涡轮增压汽油直喷（T-GDI）发动机添加剂的条件。已经积极地进行了使用ANN预测物理现象的研究，包括内燃机在内的各种应用。但是，必须在各种条件下收集大量数据才能建立这些预测。此外，对复杂现象（例如发动机燃烧过程）的预测要求在各种条件下进行数据收集。因此，在各种条件下通过实验获得这些数据非常昂贵且耗时。然而，本研究中介绍的方法可以以最少的实验数据有效预测复杂的燃烧过程，例如加氢燃烧。为了实施这种方法，目标发动机是使用商业1D发动机仿真软件GT-Power基于在选择条件下获得的某些实验结果建模的。使用GT-Power发动机模型获得了在扩展条件下进行的ANN训练的数据。根据获得的数据，构建了预测T-GDI发动机加氢燃烧过程的ANN模型，并将其结果与实验结果进行了比较。观察结果之间存在合理的一致性，证明了预测模型的有效性和可靠性。