基于神经网络的加速度时程积分方法

加速度的二重积分是位移,因此利用加速度传感器可以测量位移,但加速度信号积分中存在零点校正和边界条件确定的问题.为了准确地将加速度时程积分成对应时刻的位移,提出了一种利用非线性映射能力很强的BP神经网络建立加速度时程与位移时程之间的关系.通过对样本的学习,BP神经网络将这种非线性映射关系以分布并行的方式存储在网络的联结权矩阵中,从而对样本集进行非逻辑归纳.数值仿真结果和实测结果表明,该方法抗噪声污染能力强,收敛速度快,识别精度较高.

Time history integral method of acceleration based on artificial neural networks

The double integral of acceleration means displacement, so the accelerometer can be used to measure the displacement. Due to the difficulty in determining the zero point of acceleration and the boundary value of integration, how to convert the time history of acceleration to the time history of displacement accurately is a problem in the field of engineering. This paper proposes a method of back-propagation artificial neural networks to realize the conversion. BP networks can be used to map a set of inputs (acceleration response) to a set of outputs (displacement response) by non-linear mapping. Through learning of samples, BP networks find non-logic induction of a sample set by adjusting the connecting weights and thresholds and storing in the weight matrix. As an applicable example, the numerical value of cable displacement was attained with a relatively high precision in the wind-rain induced vibration based on BP network model. Analytical results show that BP networks are effective against noise pollution and then confirmed its validity.