基于多种神经网络的风暴潮增水预测方法的比较分析

简要介绍了利用BP神经网络、小波神经网络、递归神经网络进行风暴潮增水值预测的原理。选取广东省珠江口以南的阳江站2017年风暴潮增水数据进行测试。结果表明,三种神经网络方法针对阳江地区风暴潮增水的预测均具有可靠性和实用性。以当前增水值为输入量的单因子模型更能反映真实风暴潮增水趋势,而从增水极值预测的准确性来看,以台风风力、气压、风向等相关参数为输入量的多因子模型优于单因子模型。BP神经网络更适用于多因子长时间预测,小波神经网络在单因子短时间预测上准确性更高,递归神经网络预测值与实测值相关性更强。在工程运用中,需根据地域时空特点、数据资料的丰富度与预测值评估指标选择合适的方法。

Comparative analysis of storm surge water prediction methods based on multiple neural networks

The principle of using BP neural network, wavelet neural network and recurrent neural network to predict the storm surge water increase value are briefly introduced. The 2017 storm surge water increase data of the Yangjiang station of the Pearl River Estuary in Guangdong Province was selected for testing. The results show that the three neural network methods are reliable and practical for the prediction of storm surge water increase in Yangjiang area. The single factor model with the current water added value is more capable to reflect the true storm surge water increase trend. From the accuracy of the prediction of the maximum value of water increase, the multi-factor model with input parameters such as typhoon wind, air pressure and wind direction is better than the single factor model. BP neural network is more suitable for multi-factor long time prediction. The wavelet neural network has higher accuracy in single-factor short-time prediction, and the recurrent neural network prediction value is more correlated with the measured value. In the engineering application, it is necessary to select an appropriate method according to the spatial and temporal characteristics of the region, the richness of the data, and the predictive value evaluation index.