We have developed a hybrid machine learning (ML) model for the prediction and optimization of a gliding arc plasma tar reforming process using naphthalene as a model tar compound from biomass gasification. A linear combination of three well-known algorithms, including artificial neural network (ANN), support vector regression (SVR) and decision tree (DT) has been established to deal with the multi-scale and complex plasma tar reforming process. The optimization of the hyper-parameters of each algorithm in the hybrid model has been achieved by using the genetic algorithm (GA), which shows a fairly good agreement between the experimental data and the predicted results from the ML model. The steam-to-carbon (S/C) ratio is found to be the most critical parameter for the conversion with a relative importance of 38%, while the discharge power is the most influential parameter in determining the energy efficiency with a relative importance of 58%. The coupling effects of different processing parameters on the key performance of the plasma reforming process have been evaluated. The optimal processing parameters are identified achieving the maximum tar conversion (67.2%), carbon balance (81.7%) and energy efficiency (7.8 g/kWh) simultaneously when the global desirability index I₂ reaches the highest value of 0.65.

我们已经开发了一种混合机器学习（ML）模型，用于预测和优化使用萘作为生物质气化的模型焦油化合物的滑弧等离子体焦油重整过程。已经建立了三种著名算法的线性组合，包括人工神经网络（ANN），支持向量回归（SVR）和决策树（DT），以应对多尺度和复杂的等离子体焦油重整过程。通过使用遗传算法（GA）实现了混合模型中每种算法的超参数的优化，这表明实验数据与ML模型的预测结果之间有相当好的一致性。蒸汽碳比（S / C）是转化的最关键参数，相对重要性为38％，放电功率是决定能源效率最有影响力的参数，相对重要性为58％。评估了不同工艺参数对等离子体重整过程关键性能的耦合作用。确定了最佳工艺参数后，当全球期望指数I达到最大焦油转化率（67.2％），碳平衡（81.7％）和能源效率（7.8 g / kWh）时₂达到0.65的最大值。