基于灰色理论和神经网络的边坡位移预测

Prediction of slope displacement based on gray model and neural network

边坡位移的发展受地质条件、天气环境和人类活动等众多因素的影响,变化趋势复杂,很难建立一个准确的经典数学模型对其进行全面的描述。为了得到边坡位移较准确的估计,采用多模型信息融合技术对边坡位移进行了预测。首先,将边坡这类影响因素复杂的系统看成是一个灰色系统,分别采用GM(1,1)模型、Verhulst模型和DGM(2,1)模型对位移值进行预测。其次,考虑到神经网络的高速并行计算能力和类似人类思维活动的处理机制,利用神经网络的办法对不同的灰色预测模型进行组合,生成灰色神经网络模型,该模型有效地将灰色理论弱化数据序列波动性的优点和神经网络特有的非线性适用性信息处理能力相融合,通过反复训练、学习,自动调节,可以得出各模型在组合模型中的合理权重,从而输出满意的结果。通过对比发现,利用组合灰色神经网络模型预测的位移值,比单独的灰色模型预测的位移值具有更高的精度。

Due to the influences of various factors such as the conditions of geology, weather and environment and the activities of mankind etc. the variation trend of slope displacement is complex, and it is very difficult to create a precise representative mathematical model to make the overall description. In order to acquire the exact evaluation of the slope displacement, the multi-model information fusion technology is adopted to forecast it. At first, such system with the influences of complicated factors as slope can be regarded as a gray system, the slope displacement is respectively estimated by the GM（ 1,1 ） model, Verhulst model and DGM （2,1） model. Secondly, considering the high-speed parallel calculative activity and the disposal process similar to the mankind thinking activities, a kind of composite gray neural network forecast modeling method is put forward. The fluctuation of date sequence is weakened by the gray theory and the neural network is capable of processing non-linear adaptable information, and the model is a combination of those advantages. Based on the repeated training and learning, the reasonable weights of the neural network can be gained, and the satisfactory results can be acquired. It is more accurate of the forecasting results by the composite gray neural network model than that by the only gray models by comparison.