基于卡尔曼滤波的GA-BP模型在大坝变形预测中的应用

The Application of GA-BP Model based on Calman Filter in Dam Deformation Prediction

传统的BP神经网络拥有良好的逼近非线性映射能力,然而由于其自身存在收敛速度慢,容易陷入局部极小值和泛化能力差的不足,往往难以满足实际中预测精度的需要。采用卡尔曼滤波方法,将观测到的大坝位移原始值进行滤波处理,以尽可能剔除随机误差的干扰,并引入遗传算法,对神经网络的权、阈值进行优化,提高其全局搜索能力,建立了基于卡尔曼滤波的GA-BP模型。以某大坝位移预测为例,证明了此模型比传统的BP模型在预测精度上有所提高,具有一定的实际应用价值。

The traditional BP neural networks approximate to the nonlinear mapping. However, it has some defects, such as slow conver-gence, local minimum and bad generalization ability. It＇s often difficult to meet the needs of actual forecasting accuracy. This paper uses themethod of Caiman Filer, filtering the original data of dam displacement from observed dam so as to eliminate the disturbance of random error.And the genetic algorithm is applied to optimize the weights and thresholds of neural networks to improve global search ability. GA-BP modelis established based on Calman filter. Taking a dam displacement prediction as an example, the result shows that this model has been im-proved over the traditional BP model in forecasting precision and has certain application value.