基于机器学习的鱼雷推进控制用镁海水电池性能预测  

Based on Machine Learning Seawater Batteries of Performance Prediction for Torpedo Propulsion Control

作　　者：刘笑达[1,2] 侯建鹏 阴明[1] 王志伟 侯利锋[1] 卫英慧[1,3] LIU Xiao-da;HOU Jian-peng;YIN Ming;WANG Zhi-wei;HOU Li-feng;WEI Ying-hui(Taiyuan University of Technology,Taiyuan 030024,China;North Automatic Control Technology Institute,Taiyuan 030006,China;Taiyuan University of Science and Technology,Taiyuan 030024,China)

机构地区：[1]太原理工大学,太原030024 [2]北方自动控制技术研究所,太原030006 [3]太原科技大学,太原030024

出　　处：《火力与指挥控制》2020年第9期52-57,62,共7页Fire Control & Command Control

基　　金：山西省自然科学基金资助项目(201901D111102)。

摘　　要：针对鱼雷用镁海水电池阳极放电性能低以及传统“试错法”在材料设计中导致开发周期过长的问题。通过数据分析和机器学习的方法,采用线性回归(Linear Regression,LR),支持向量回归(Support Vector Regression,SVR)和神经网络(Multilayer Perceptron,MPL)算法对数据集进行训练建立模型,使用预测模型对镁基阳极材料的放电性能进行预测,根据预测结果制备了Mg-5.7Al-0.9Ge合金作为镁海水电池用阳极材料。最后,通过电化学实验对Mg-5.7Al-0.9Ge合金在3.5 wt%NaCl溶液中的放电性能进行验证研究,研究发现该合金分别在20 mA·cm^-2、50 mA·cm-2电流密度下,放电电位分别为-1.641 V和-1.429 V,放电效率分别为69.5%和60.4%,其放电性能优于商用镁合金阳极材料AZ61。结果表明,SVR算法建立的模型预测能力最佳,具有较高的相关系数和较低的误差,为镁基阳极材料的成分设计和快速开发问题提供指导。Aiming at the problem of low anode discharge performance of magnesium seawater batteries for torpedoes and the traditional"trial and error method"in the material design cause the development cycle is too long.Through data analysis and machine learning methods,Linear Regression(LR),Support Vector Regression(SVR)and Multilayer Perceptron(MLP)algorithms are used to model the data set,and the discharge performance of magnesium-based anode materials is predicted by the model.According to the prediction results,Mg-5.7Al-0.9Ge alloy was prepared as anode material for magnesium seawater battery.Through electrochemical experiments to verify the discharge performance of Mg-5.7Al-0.9Ge alloy in 3.5 wt%NaCl solution,the study found that the alloy was under the current density of 20 mA·cm-2,50 mA·cm-2,the discharge potential is-1.641 V and-1.429 V,and the discharge efficiency is 69.5%and 60.4%,respectively,and its discharge performance is better than the commercial magnesium alloy anode material AZ61.The results show that the SVR model has the best predictive ability,high correlation coefficient and low error,and provides guidance for the composition design and rapid development of magnesium-based anode materials.

关 键 词：镁海水电池 机器学习 支持向量回归 镁基阳极材料 放电性能 

分 类 号：TM912[电气工程—电力电子与电力传动] TJ63[兵器科学与技术—武器系统与运用工程]