基于BP神经网络的精密温度控制

BP Neural Network for High-Precision Temperature Control

为解决在精密温控系统中的传统比例-积分-微分(Proportion Integration Differentiation,PID)控制参数整定复杂、超调严重等问题,设计并搭建了精密温控系统,并且系统基于随机梯度下降(Stochastic Gradient Descent,SGD)准则,采用反向传播(Back Propagation,BP)神经网络算法实时动态调整PID参数。实验结果表明,BP神经网络PID温控系统具有更小超调和更快收敛速率等优势,系统超调量与弛豫时间分别比传统PID控制系统减少13.35%和18.56%。同时,在6 h测量时间内,被控对象温度波动的标准偏差达到0.00036℃。

Aiming at the problems of complex parameters setting and serious overshoot of classical Proportion-Integration-Differentiation(PID)algorithm,a high-precision temperature control scheme based on the Stochastic Gradient Descent(SGD)criterion and the Back Propagation(BP)neural network algorithm is designed and built,with the ability of dynamically adjusting the PID parameters in real-time.The experimental results show that the proposed temperature control system based on BP neural network PID has the advantages of smaller overshoot and faster convergence rate.The overshoot and relaxation time of the new system are reduced by 13.35%and 18.56%,respectively.Meanwhile,the standard deviation of temperature fluctuation reaches to 0.00036℃during the 6 h measurement time.