基于神经网络的传感器动态补偿算法及DSP实现

摘要

In order to improve the dynamic response characteristics of the sensor, dynamic compensation for its output is an effective method. The dynamic sensor inverse modeling method based on adaptive neural network is discussed, the Network partition training and variable study factors is adopted for improving training of variable precision and shortening the time of convergence. The feasibility of dynamic compensation method was tested in different SNR of random noise. Simulation results from typical piezoelectric model show that the sensor has ideal dynamic response characteristics after being compensated, the noise are suppressed. Then hardware of data acquisition and compensation system is designed based on DSP, the sampling data from simulator of sensor show that the system not only samples and stores data accurately, but also corrects the dynamic errors caused by simulator of sensor.%为了改善传感器的动态响应特性.对其输出结果进行动态补偿是一个有效方法;讨论了基于自适应神经网络的传感器动态逆建模方法,采用网络分块训练和可变学习因子的方法来提高训练的精度,缩短收敛时间;研究了在加入不同信噪比的随机噪声下应用该模型实现传感器动态补偿的可行性;对典型的压电传感器模型进行了仿真,仿真结果表明补偿后传感器模型的响应速度加快,同时还可以抑制噪声;研制了基于数字信号处理器的数据采集及补偿系统并运用该系统对传感器模拟器输出的数据进行了采集,试验结果表明该系统能够准确的采集存储数据,同时还能够修正由传感器模拟器引起的动态误整.