一种优化的GA-KF与BP-Adaboost地表下沉组合预测模型北大核心CSCD

为提高矿区GNSS CORS自动化监测系统地表下沉预测精度,提出一种结合小波分析,采用遗传算法优化的卡尔曼滤波模型(GA-KF)与相空间重构的BP神经网络强预测模型(BP-Adaboost)的组合预测方法。利用小波分析获取原始监测数据的趋势项和随机项,并分别通过GA-KF模型和相空间重构BP-Adaboost模型预测趋势项和随机项,两者之和即为最终预测结果。以亳州板集矿监测站数据为例进行预测,结果表明:1)与单一使用GA-KF和相空间重构BP-Adaboost模型预测值对比,该组合模型预测精度更高;2)组合模型受建模序列长度影响较小,平均相对误差在0.003%以内,远小于两种单一模型,具有一定抗干扰性。

An Optimized Combined Prediction Model for Surface Subsidence Based on GA-KF and BP-Adaboost

To improve the prediction accuracy of surface subsidence of the GNSS CORS automatic monitoring system in mining areas, we propose a combined prediction method based on the Kalman filter model optimized by genetic algorithm (GA-KF) and the BP neural network strong prediction model (BP-Adaboost) after phase space reconstruction based on the wavelet analysis. The trend sequence and random sequence of the original monitoring data are obtained by wavelet analysis, and are predicted by the GA-KF model and the BP-Adaboost model respectively. The sum of the two data is the final prediction result. Taking the data of Bozhou Banji monitoring station as an example, the experimental results show that: 1) Compared with the prediction values of using the single GA-KF and BP-Adaboost model after phase space reconstruction, the combined model has higher prediction accuracy. 2) It is found that the combined model is less affected by the length of modeling sequence, and the average relative error is within 0.003%, which is much smaller than the two single models and has a certain anti-interference ability.