基于机器学习的便携式液体密度仪设计

Design of portable liquid density meter based on machine learning

蓄电池是储能电站中必不可少的重要组成模块,而该类电池在生产过程中电解液的相对密度是影响蓄电池性能的重要因素。为了满足密度快速准确的测量需求,基于机器学习开发了一套便携密度实时测量装置。基于密度测量谐振理论,讨论并分析了液体温度与谐振频率的关系,区别于现有单点测温的多项式拟合技术,采用了全域多点温度场、谐振频率和密度特征向量构建并训练神经网络,以实现更高精度的密度测量。试验结果表明,本设备可很好地消除温度对密度测量时的影响,平均绝对误差约为1×10^(-4)g/cm^(3)。相较于目前基于多项式拟合的传统方法(1×10^(-3)g/cm^(3)),测量误差最多减少了约1个数量级,单次测量时间少于0.1 s,可满足电解液密度的实时高精度测量需求。

The battery is an essential component module in the energy storage power station,and the relative density of the electrolyte in the production process of the battery is an important factor affecting the performance of the battery.In order to meet the need of fast and accurate density measurement,this paper develops a portable density real-time measurement device based on machine learning.Based on the resonance theory of density measurement,the relationship between liquid temperature and resonance frequency is discussed and analyzed.Different from the existing polynomial fitting technique of single point temperature measurement,the global multi-point temperature field,resonance frequency and density eigenvector are used to construct and train the neural network to achieve higher precision density measurement.The experimental results show that the device can eliminate the influence of temperature on density measurement,and the average absolute error is about 1×10^(-4)g/cm^(3).Compared with the traditional method based on polynomial fitting(1×10^(-3)g/cm^(3)),the measurement error is reduced by about 1 order of magnitude,and the single measurement time is less than 0.1 s.It can meet the requirements of real-time and high-precision measurement of electrolyte density.