基于改进RBF网络的传感器非线性误差补偿

Nonlinear error compensation of sensors based on improved RBF network

为推动胃肠道动力功能障碍型疾病临床诊查技术的发展,研制了胃肠道多元生理参数无创检测系统,针对该系统中压力传感器的非线性误差补偿问题进行研究。介绍了系统所采用的扩散硅压阻式绝对压力传感器的原理,分析了这类传感器的非线性误差产生原因。在传统的减法聚类算法的基础上,提出基于改进的减法—密度聚类算法的RBF网络的传感器非线性误差补偿方法,对样本数据进行聚类操作,用来确定RBF神经网络的初始聚类中心,并结合梯度下降法对网络参数和权值进行训练。结合实际系统的实验数据进行了方法验证和效果分析。实验结果表明:方法在系统误差纠正方面比传统方法提高至-1～4 kPa,使得测量结果准确性得以较大的提高,满足了系统的应用需求。

In order to improve the development of gastrointestinal dyskinesis clinical detecting technology,a detecting system of gastrointestinal multiple biological parameters is established and the error compensation for its pressure sensor＇s non-linear error is investigated.The basic theory of the system＇s spread silicon piezo-resistance absolute pressure sensor is presented and the reason of the sensor＇s non-linear error is analyzed.Based on traditional substractive clustering algorithm,an improved substractive-density clustering method is presented to form a RBF neural network for the sensor＇s non-linear error compensation.It is used to cluster samples for network＇s initial centers.Gradient descent algorithm is applied to train the network.The method＇s effectiveness and practicability are proved by the experiment with actual system＇s test data.The result indicates that the system＇s error is corrected into the region from-1 to 4 kPa,which is better than traditional methods.The system is optimized,the performance is improved and the system＇s practical need is satisfied.