自蔓延高温合成多孔NiTi合金孔隙的SVR预测

根据自蔓延高温合成法(SHS)制备多孔NiTi合金孔隙试验所获得的实测数据集，应用基于粒子群算法(PSO)寻优的支持向量回归(SVR)方法，建立不同反应参数(温度，粒度和压坯密度)下合成的多孔NiTi合金孔隙的SVR预测模型，并与基于误差反向传播神经网络(BPNN)回归模型的预测结果进行比较。结果表明：在相同的训练与测试样本集下所获的SVR预测结果的平均绝对百分误差(MAPE)比BPNN预测模型的要小，其预测精度更高，预测效果更好；SVR-LOOCV预测的MAPE也比BPNN略小，且其预测结果的相关系数达到了0.999。因此，该方法是一种预测SHS法制备多孔NiTi合金孔隙的有效方法，可为SHS合成多孔NiTi提供理论指导

Support Vector Regression Prediction of Porosity of Porous NiTi Alloy by Self-Propagation High-Temperature Synthesis

Based on the experimental dataset, the support vector regression (SVR) combined with particle swarm optimization (PSO) for parameter optimization, is proposed to establish a model for estimating the porosities of NiTi alloys synthesized by self-propagation high-temperature synthesis (SHS) approach under different process parameters, including temperature, particle size and green density. The prediction results indicate that the mean absolute percentage error (MAPE) achieved by SVR is smaller and more accurate than that of back-propagation neural network (BPNN) for identical training and test samples, reflecting the prediction ability of SVR is superior to that of BPNN; MAPE predicted by leave-one-out test of SVR (SVR-LOOCV) is also slightly better than that of BPNN, and the correlation coefficient (R2) reaches 0.999. Therefore it is demonstrated that SVR is a promising and practical technique to estimate the porosity of porous NiTi alloy synthesized under different SHS process parameters, and can provide a reasonable guidance for the SHS of porous NiTi theoretically