基于人工神经网络的海底细粒土工程分类研究

Study on Engineering Classification of Fine Submarine Soil Based on Artificial Neural Network

海洋沉积物工程定名对于开展海洋工程建设具有重要作用,然而海底粉土和黏性土的定名受人为因素影响容易产生误差。使用人工神经网络的方法对黄河口埕岛海域284组细粒土数据进行了训练和学习,得到了只利用沉积物粒径质量分数进行定名的方法。结果表明,使用人工神经网络的方法能够有效地对沉积物进行工程定名。当网络含有5个输入层节点、9个隐藏层节点、3个输出层节点、训练函数为Scaled conjugate gradient时定名准确率最高,检验准确率高达97.7%。训练数据的数量是造成神经网络预测存在误差的重要因素,随着数据量的增加,网络的可靠性和通用程度将越来越高。

The engineering naming of marine sediments plays an important role in the development of marine engineering construction.However,the naming of silt and clay in the seabed is easy to be affected by human factors.Using the artificial neural network method,this study trained 284 sets of fine-grained soil data in Chengdao sea area of the Yellow River estuary and proposed a naming method using granularity data.Results showed that the method of artificial neural network performs well in fine submarine soil engineering naming.The accuracy of net naming is the highest,which as high as 97.7%,when the network contains 5 input layer nodes,9 hidden layer nodes,3 output layer nodes,and scaled conjugate gradient as the training function.The number of training data is an important factor that causes errors in neural network prediction.With increasing data volume,the reliability and universality of the network will become higher and higher.