基于BP神经网络的微量药品动态称重系统非线性补偿  

Nonlinear compensation of micro scale capsule dynamic condition weighing unit based on BP neural network model

作　　者：庄育锋[1] 胡晓瑾[1] 翟宇[1] 

机构地区：[1]北京邮电大学自动化学院,北京100876

出　　处：《仪器仪表学报》2014年第8期1914-1920,共7页Chinese Journal of Scientific Instrument

摘　　要：针对微量药品动态称重系统中电阻应变式称重传感器的输出电压与药品单元质量之间的非线性关系问题,提出了基于BP神经网络的非线性补偿方案。基于L-M算法建立了BP神经网络模型,实现了电阻应变式称重传感器的输入与输出非线性补偿校正,并与bfgs拟牛顿算法、Scaled共轭梯度算法所建立的BP神经网络模型对比,重点比较了模型预测输出、误差性能分析、回归分析。仿真实验结果表明:基于L-M算法建立的BP神经网络模型,在收敛速度、误差性能方面具有更高效的表现,有利于微量药品动态称重系统中称重传感器的非线性特性的有效校正。Aiming at the nonlinear characteristic between the weighing sensor output and the weight of capsule unit in micro scale capsule dynamic weighing system,a nonlinearity compensation scheme based on BP neural network is proposed.A BP neural network model is established based on Levenberg-Marquardt algorithm.The model implements the nonlinearity compensation between the output voltage of weighing sensor and the input of capsule unit weight.The proposed method was compared with bfgs quasi-Newton algorithm and scaled conjugation gradient algorithm,and the model performances of forecasting output,error performance analysis and regression analysis were compared.Simulation results show that the BP neural network model based on Levenberg-Marquardt algorithm has high performance in terms of convergence rate and error performance.The model is more suitable for the nonlinearity compensation in micro scale capsule dy-namic weighing system.

关 键 词：微量 药品称重 动态 BP神经网络 LEVENBERG-MARQUARDT算法 拟牛顿算法 Scaled共轭梯度算法 误差性能分析 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置] TP274[自动化与计算机技术—控制科学与工程]