基于MEC优化BP神经网络的PSD非线性校正

光电位置敏感传感器(PSD),特别是其B区存在非线性误差大、测量精度低的问题.针对现有神经网络校正方法的不足,提出一种基于思维进化计算(MEC)算法优化的神经网络校正模型.该方法首先应用MEC算法搜索最优神经网络初始权值和阈值,再利用LM算法训练BP神经网络,最后将训练好的神经网络用于PSD非线性校正.仿真实验结果表明,所提出的方法校正精度高,收敛速度快,泛化能力强,测试数据的平均误差被控制在0.005 mm以下.经过校正后的PSD在非线性区表现出与线性区相似的线性程度,提高了PSD的测量精度.

Nonlinear Correction of PSD Based on MEC Optimized BP Neural Network

Position sensitive detector (PSD),especially its B region,has the problems of large nonlinear error and low measurement accuracy.To overcome the deficiencies of existing neural network methods,an optimized neural network nonlinear correction model based on MEC is proposed.First,the MEC algorithm is adopted to search the optimal initial weights and thresholds of the BP neural network,then,the Levenberg Marquardt algorithm is adopted to train the BP network.Finally,the trained network is used for PSD nonlinear correction.The simulation results show that the proposed correction method has high correction accuracy,fast convergence speed and strong generalization ability.The average test error is less than 0.005 mm.The linearity of corrected PSD in the nonlinear region is similar to the linear region,therefore,the measurement accuracy of PSD is improved.