基于人工神经网络的三峡库区丰都县水库塌岸预测

为了建立一个适合于三峡库区的塌岸预测方法体系，采用具有处理非线性关系功能的人工神经网络方法对水库塌岸问题进行研究。通过训练、学习和仿真，获得预测正确率为97.2%的具有7-32-14网络结构的BP神经网络模型，采用该模型对蓄水位为175　m时丰都县各岸段进行塌岸预测，并将预测结果与传统经验公式计算法所得结果及实际监测数据进行对比。结果表明:基于人工神经网络的塌岸预测宽度与实际监测数据很接近，偏差在5 m以内；公式法计算结果与监测值平均偏差为15.9 m，而且对于部分坡段，公式法计算结果比实际监测值小8～11 m，没能预测出塌岸的真正范围。采用神经网络模型对丰都县水库进行塌岸预测，预测结果与实际监测数据平均偏差约3.8%，表明其预测结果可靠。

Reservoir Bank Collapse Prediction of Fengdu Couty in Reservoir Area of Three Gorges Based on Artificial Neural Network

In order to establish an appropriate methodology of bank collapse prediction in the Three Gorges reservoir area, the article applied the artificial neural network method to study the bank collapse prediction. The method has the function of disposing the nonlinear relation. Through training, learning and simulation, it obtained the BP neural network model with a 7-32-14 network structure, and the correct prediction rate of it was 97.2%. Then  the model was used to forecast the reservoir bank collapse of Fengdu county when the water level was 175 m. Compared the forecast results with the traditional empirical formula calculation and the actual monitor data. The comparative results showed that the bank collapse prediction based on artificial neural network were very similar to the monitor data, and the deviation was less than 5 m. The average deviation of the formula calculation results and the monitor data was 15.9 m. For part of the slope sections, the formula calculation results were 8-11 meters,which was less than the actual monitor data. It failed to predict the true scope of bank collapse. In short, the average deviation of the bank collapse prediction results using artificial neural networks is about 3.8%. It is reliable and has higher precision.