基于自适应RBF神经网络的连续压力波信号滤波方法

Filtering Method for Continuous Pressure Wave Signal Based on Adaptive RBF Neural Network Algorithm

由于钻井液连续压力波信号与正脉冲压力信号检测原理不同,致使固定参数滤波方法存在检测特征点时间不准确、误码率高等问题。针对该问题提出一种小波包变换结合自适应变步长RBF神经网络非线性滤波器的滤波方法。该方法首先对输入的连续压力波信号进行小波包变换,运用分层阈值滤波算法和奇异值分解算法,分离出含噪声的有用连续压力波信号;对输入的不发码信号进行带通滤波,分离出噪声相关信号。然后将上述两路信号输入RBF神经网络中,通过自适应变步长滤波算法进行滤波处理,输出有用连续压力波信号。仿真结果表明:该滤波方法与固定参数滤波方法相比,滤波后信号与原信号的相关系数、均方误差、信噪比都得到了提升。现场应用中,相比固定参数滤波算法误码率降低10%,连续压力波信号的噪声得到有效抑制。

Because the detection principle of continuous pressure wave signal of drilling fluid is different from that of positive pulse pressure signal,the traditional fixed parameter filtering method does not accurately detect the time of characteristic point and bit error rate is high for continuous pressure wave signal of drilling fluid.To solve this problem,a filtering method based on wavelet packet transform and adaptive variable step RBF neural network nonlinear filter is proposed.Firstly,the wavelet packet transform of the input continuous pressure wave signal is carried out,and the useful continuous pressure wave signal with noise is separated by using the layered threshold filtering algorithm and singular value decomposition algorithm;The input non code signal is filtered by band-pass,and the related noise signal is separated.Then,the above two signals are input into RBF neural network and filtered by adaptive variable step filtering algorithm to output useful continuous pressure wave signal.The simulation result shows that:compared with the fixed parameter filtering method,the correlation coefficient,mean square error and signal-to-noise ratio of the filtered signal using the proposed filtering method to the original signal are improved.In the field application,compared with the fixed parameter filtering algorithm,the bit error rate is reduced by 10%,and the noise of continuous pressure wave signal is effectively suppressed.