压力传感器的校正的改进超限学习机方法

摘要

With the development of test technology, pressure sensors are widely used in various fields of precision measure-ment and test.Due to the effects of the measured object, environment and other factors, the input /output characteristics of the sensors have many kinds of errors, yet the compensation method using the commonly neural network cannot solve the problem well.In order to overcome the limitation in nonlinear correction of sensors, an approach based on wavelet function and Extreme Learning Machine ( ELM) was presented.The nonlinear correction principle of sensors was explained.And the realization process of wavelet Extreme Learning Machine ( WELM) was introduced.A pressure sensor was adjusted with BP neural network,tradi-tional ELM and the method of WELM, respectively.The experiment results show that BP neural network reduces the maximum relative fluctuation from the initial 22.32%to 1.758%,traditional ELM reduces the maximum relative fluctuation to 0.038%.Mo-reover, the method of WELM reduces the maximum relative fluctuation to 0.026%, Therefore it evidently improves the perform-ance of the sensor and achieves better result.Meanwhile,the training error curves show that convergence rate and training error of WELM are significantly better than BP neural network.%随着测试技术的发展，压力传感器被广发的应用于各种精密测量和检测领域。由于受到被测量对象、测量环境等因素影响，其输出会产生各种误差，而常用的神经网络不能很好解决此问题。文中提出一种采用小波函数和超限学习机相结合的新方法，阐述了传感器的非线性校正原理和小波函数超限学习机的实现过程，并分别采用BP 神经网络法，传统超限学习机和小波超限学习机对压力传感器进行非线性校正。实验结果表明：BP 神经网络法使得传感器的最大相对波动由初始的22．32％降低到1．758％；而传统超限学习机方法使其降低到0．038％，小波超限学习机则使其降低到0．026％，改善了传感器的工作性能；同时训练过程误差曲线图表明小波超限学习机的收敛速度和训练误差，显著高于BP神经网络。