差动传感器残余电压及非线性的神经网络补偿

Neural Network Compensation of Residue Voltage and Nonlinear for Differential Sensor

为消除差动变压式传感器零点残余电压及非线性特性,提出基于重新参数化的B样条函数以及粒子群算法β参数B样条神经网络(B-BP-PSO)。由粒子群算法(PSO)取代传统BP算法,并由其搜索最佳β因子,用以得到适合本网络权值搜索的最优重新参数化B样条基函数,从而使得该神经网络可有效克服传统算法易于陷入局部最优的缺点。实验结果表明:经其校正后的差动变压式传感器的最大输出误差为11 mV,最大相对误差为3.7%,零点残余电压为5 mV.该方法可有效消除各种参数对差动变压式传感器输出结果的零点残余电压及非线性影响,且可用于其他各类传感器的非线性校正,具有很大的实际应用价值。

To eliminate the zero-point residue voltage and nonlinear characteristic of differential transformer sensor,the β parameterized B-spline neural network（B-BP-PSO）based on re-parameterization B-spline function and particle swarm algorithm was proposed.Traditional BP algorithm was replaced by PSO to find the best suitable β factor,thereby,receive the best re-parameterization B-spline basic functions fit for searching weights of network which make the neural network can overcome the shortcomings of easily going into local optimum.Experimental results show that after calibration,the maximum output error of the differential transformer sensor is 11 mV,maximum relative error is 3.7%,zero-point residue voltage is 5 mV,so the strategy can effectively eliminate the zero-point residue voltage and nonlinear influence by various parameters,it is also practicable for other type of sensor nonlinear correction and has great practical value.