基于BP神经网络的表层悬浮物浓度预测模型

Modeling with BP Neural Network for Predicting Suspended Solid Concentration in Surface

悬浮物浓度是海洋沉积动力学领域中的重要参数,对其进行准确预测及定量研究绿潮爆发期间悬浮物所带来的影响具有重要意义。将BP(Back Propagation)神经网络应用于表层悬浮物浓度的预测中:将流速、水深、波高、温度、盐度及风速等影响悬浮物浓度的因素作为BP神经网络的输入单元,通过对苏北近岸海域进行调查,获取用于训练和预测的数据,建立表层悬浮物浓度的BP神经网络预测模型。将预测结果与多因子逐步回归拟合结果进行比较,得到逐步回归预测结果的平均相对误差为24.13%,BP神经网络预测结果的平均相对误差仅为13.02%。由此可见,BP神经网络预测结果具有更高的精度,可为苏北近岸海域表层悬浮物浓度的准确预测提供更可靠的途径。

Accurate prediction of suspended solid concentration is very important to quantitative study on the impact of the suspended solids during green tides.The BP neural network model for predicting suspended solid concentration surface is established and trained by the data acquired from SuBei Coastal Waters,with the input consisting of velocity,water depth,wave height,temperature,salinity and wind speed.Trials indicate that the relative error of the average prediction of the BP neural network is 13.02%,in contrast to 24.13% of that from the multi-factor stepwise regression.