钱塘江涌潮传播时间预报方法研究

Research on Forecast Method of Tidal Bore Traveling Time in Qiantang River

群众观潮、防潮管理、交通航运、涉水工程建设等均需及时、准确、可靠的钱塘江涌潮预报信息,而经验预报方法和常规传统机器学习算法未全面考虑各种影响因素对钱塘江涌潮传播时间的影响,预报精度无法保证。对此,综合考虑天文因素、风力风向、上游来水、江道地形、前日传播时间等主要影响因素对涌潮的影响,构建特征集作为输入矩阵,基于卷积神经网络和全连接神经网络构建钱塘江涌潮传播时间预报模型,通过粒子群算法得到最优参数,建立CNN最佳预报模型,同时基于支持向量机、BP神经网络算法建立预报模型。钱塘江仓前至七堡站涌潮传播时间2009~2017年预报结果表明,卷积神经网络预报模型大中潮汛预报误差在5、10 min内的比例分别高于72%、96%,预报精度高于支持向量机、BP神经网络预报模型。

Timely, accurate and reliable tidal bore forecast information of Qiantang River is required for tiding sightseeing, moisture-proof management, transportation and shipping, and wading engineering construction. However, the empirical forecast method and conventional machine learning algorithm can not fully consider the impact of various influencing factors on traveling time of Qiantang River tidal bore, and the forecast accuracy can not be guaranteed. Considering the impact of astronomical factors, wind power, upstream inflow, river channel topography, traveling time of the previous day and other main influencing factors on tidal bore, the feature set was constructed as the input matrix. Based on convolution neural network and fully connected neural network, the forecast model of tidal bore traveling time in Qiantang River was established. The CNN optimal model was obtained by optimizing the model parameters based on particle swarm optimization algorithm. At the same time, the forecast model was established based on support vector machine and BP neural network algorithm. Finally, the analysis of forecast examples of tidal bore traveling time forecast from Cangqian to Qibao station in Qiantang River from 2009 to 2017 shows that the proportion of forecast error of convolution neural network forecast model during spring and medium tide within five minutes and ten minutes is higher than 72% and 96%, and the forecast accuracy is higher than that of support vector machine and BP neural network forecast model.