基于人工神经网络的卡尔曼滤波实时校正技术

The Kalmann filter real-time adjustment technique based on artificial neural networks

针对人工神经网络技术在实际应用中常出现的过拟合现象 ,设计了以人工神经网络模型做初级预报 ,用卡尔曼滤波技术对初级预报结果进行二次预报的方法。该方法用于淮河王家坝水文站最高洪水位的预报和岷江上游段紫坪埔水文站的流量预报 ,并与标准的BP网络模型以及卡尔曼滤波模型进行了比较。两个应用实例的计算结果表明 ,以上两种技术的结合 ,不仅有利于预防过拟合问题 。

In the light of the problem of over-training often countered in the application of artificial neural networks, this paper puts forward a method of using artificial neural network model to make pre-forecast and then using the Kalmann filter real-time adjustment model to do the second forecast for the results of pre-forecast. This method has been applied to the highest flood level forecast in the Wangjiaba hydrological station in the Huaihe River and to the discharge forecast in the Zipingpu hydrological station in the upper reach of the Mingjiang River and has been compared with the results from the standard BP network model and the Kalmann filter model. The calculation results in such two application examples showed that the combination of such two techniques is not only useful to protect the over-training problem but also can enhance forecast accuracy.