基于BP神经网络的自适应Kalman滤波在滑坡沉降监测中的应用研究

Application of Adaptive Kalman Filter Based on BP Neural Network in Landslide Settlement Monitoring

利用BP神经网络对自适应Kalman滤波估值残差值进行预测,进而对滤波估值进行修正。将自适应Kalman滤波得到的状态估值、状态一步预测值、增益矩阵以及新息向量作为输入,滤波估值的残差值作为输出设计BP神经网络。对某滑坡沉降监测资料处理和分析得到基于BP神经网络的自适应Kalman滤波预测残差在-0.42mm到0.072mm之间,相较于自适应Kalman滤波预测精度和稳定性均得到显著提高,说明该方法在实际项目中具有很高的适用性。

BP neural network is used to predict the residual value of adaptive Kalman filter estimation, then the filter estimation is corrected. The state estimation, the state one-step prediction, the gain matrix and the new interest vector which are obtained by the adaptive Kalman filtering are taken as the input and the residual value of the filtering estimation is used as the output design BP neural network. Analysis of the monitoring data of a landslide settlement obtains that the prediction residual of adaptive Kalman filter based on BP neural network is between -0.42 mm and 0.072 mm, which is significantly improved compared with the adaptive Kalman filter prediction accuracy and stability, indicating that the method in the actual project has high applicability.