基于STL的海表面温度预测算法

海表面温度是海洋科学研究中重要的参数之一,有效预测海表面温度对海洋灾害预警、海洋经济以及海洋生态环境研究具有重大意义。针对海表面温度具有周期性、持续性、非平稳性和非线性的特性,首先利用基于局部加权回归的周期趋势分解方法将原始海表面温度序列分解为周期项、趋势项和余项,挖掘海表面温度的潜在信息并去除序列中的随机噪音,再结合长短期记忆网络模型的优点,搭建神经网络来预测未来5天内的海表面温度。通过与其它模型的预测效果进行对比,实验结果表明,本文方法在预测海表面温度时具有较好的预测精度,能够实现海表面温度的有效预测。

Sea surface temperature prediction algorithm based on STL model

Sea surface temperature is one of the important parameters in marine scientific research, and effective prediction of sea surface temperature is of great significance for marine disaster warning, marine economy and marine ecological environment. Aiming at the characteristics of periodicity, persistence, non-stationarity and non-linearity of sea surface temperature, firstly, the original sea surface temperature series is decomposed into periodic items, trend items and residual items by using the Seasonal-Trend decomposition procedure based on Loess to mine the potential information of sea surface temperature and remove the random noise in the sequence. Combining the advantages of the long short-term memory network model, a neural network is model to predict the sea surface temperature in the next five days. Comparing with the prediction effects of other methods, the experimental results show that the proposed method has better prediction accuracy when predicting sea surface temperature and can effectively predict sea surface temperature.