机器学习在储能陶瓷Ba(Ti_(1-x)Hf_x)O_3介电常数寻优中的应用

Application of Machine Learning in Optimization of High-permittivity Energy-storage Ba(Ti_(1-x)Hf_x)O_3 Ceramic

为了加速开发具有高介电常数高储能密度的三临界电介质的过程,在Ba(Ti_(1-x)Hf_x)O_3陶瓷高介电常数成分设计中采用机器学习寻优方法与实验制备表征互相迭代的方法,建立了适用于寻找三临界成分的机器学习模型,通过机器学习寻优加速寻找具有高介电常数的三临界成分,并对几种不同优化算法的寻优效率和收敛速度进行了比较。结果表明:设计出的的高介陶瓷(摩尔分数x=11%)相对介电常数最大值约为4.5×10~4,高于传统陶瓷材料(约为1 000);同时有效地减少了尝试实验次数(约37.5%)。研究表明机器学习寻优方法可以为新型高介、高储能密度陶瓷材料开发提供新的设计方法。

In order to accelerate the design process of high dielectric permittivity materials, the machine learning opti- mization iterating with fabrication and experiment characterization method was employed in designing the high dielectric permittivity tricritical point Ba（Ti1-xHfx）O3 ceramic. During the process, the optimization machine learning model was built to accelerate the searching for high-permittivity tricritical point, and several possible algorithms＇ efficiency and con- vergence rate have been compared and discussed. The results show that the largest relative permittivity is found to be 4.5×10^4 at the composition of x=11%, which is much higher than that of normal ceramics （about 1 000）; and the efficien- cy has been improved by 37.5%. This finding may provide a new method for designing high permittivity and energy density ceramics dielectrics.