PSD非线性误差校正研究

Study on PSD non-linear error correction

为了提高遥控导弹攻击的精确度,针对影响其精确度的主要因素是半导体光电位置敏感器件(position sensitive detector,PSD)的非线性,提出基于改进型BP神经网络算法校正某遥控导弹的PSD非线性误差方法。利用LM改进BP算法以及BP算法训练数据时,使用两个隐含层对神经网络进行训练,结果为:两个隐含层的神经元个数分别为38和34;网络的第一、第二隐含层以及输出层采用的激励函数分别为tansig、tansig、logsig;编程设定最大训练次数为500,目标收敛精度为1×10-4;设置训练函数为trainlm。利用未经训练的数据对网络进行测试,该网络的计算输出误差大约在0.01 mm之内,其中最大误差为0.015 mm。理论分析与仿真结果表明,采用该方法后,即使目标发生机动,遥控导弹也能正确攻击到目标。

In order to improve the attack accuracy of remote control missiles, according to the main factor influencing the accuracy （the PSD nonlinear） and based on improved BP neural network algorithm, the PSD nonlinear error correction method of one remote control missile was proposed. BP algorithm and training data were improved using LM, two hidden layers were used to train the neural network training and the results as follows： the neurons number of the two hidden layers are respectively 38 and 34; the excitation functions of first and second hidden layer of the network and output layer are respectively tansig, tansig and Iogsig; maximum training number of programming is 500; the target convergence precision is 1 ×10^-4; the training function is trainlm. Using the untrained data to test the network, the calculation output error is about 0.01 mm, and the maximum error is 0.015 mm. Theoretical analysis and simulation results show that after using the proposed method, even if the target happens flexibility, remote control missile can correctly attack the target.