摘要
为了提高非线性时序预测模型的精度,利用自相关技术分析了海洋冰情时序的延迟特性,据此确定了RBF网络的输入、输出向量,给出了MATLAB环境下海洋冰情预测的高精度径向基函数(RBF)神经网络的结构、设计、仿真函数和图形结果的输出方法,建立了海洋冰情预测的高精度RBF网络模型。使用27年的海洋冰情实测资料进行了网络的训练和检验,并将之用于预测,各训练样本的误差为0.0,预测值的精度高于门限自回归模型预测的精度。实例分析表明,所构建的RBF网络模型能充分利用预报因子的信息和神经网络方法的非线性映射能力,模型稳定性好,精度高,可广泛应用于各种自然灾害的非线性时序动态预测。
In order to raise the precision of sea ice condition prediction model, a high precision MATLAB radial basis function (RBF) network model of sea ice condition prediction is given. The structure, design, simulation and figure output of this model are developed. The delay time of sea ice condition time series is analyzed with auto-correlation technique for the input and output of this MATLAB radial basis function network. And a radial basis function network model for prediction sea ice condition is tested in case study. Then the training and test of the network is carried out with the recorded data of sea ice condition for 27 years. The result shows that the error of every training sample is 0.0 and the precision of the predicted value from this new model is higher than that of threshold auto-regression model. This RBF network model is stable and high precise due to its adequate utilization of relevant information on predicated factors and nonlinear mapping ability of the neural network method. And it is a good nonlinear prediction model for various natural disasters.
出处
《自然灾害学报》
CSCD
北大核心
2004年第4期105-108,共4页
Journal of Natural Disasters
基金
国家重点基础研究发展规划项目(G1999043605)
关键词
冰情
时间序列
非线性预测
径向基函数网络
精度
ice condition
time series
nonlinear prediction
radial basis function network
precision