针对气象数据多为时间序列, 而传统预测方法没有将时间相关性考虑在内, 导致预测准确率低的问题, 提出一种基于随机森林的长短期记忆网络气温预测模型.利用随机森林选择出与气温高度相关的气象要素作为输入变量, 消除原始气象数据中的噪音、降低网络的复杂度, 在此基础上利用长短期记忆网络建立总体预测模型, 在采集的多要素气象数据上进行实验.实验结果表明, 该模型在处理大规模多变量的时间序列数据时具有较高的预测精度和较强的泛化能力.
Aiming at the problem that meteorological data are mostly time series, while traditional prediction methods do not take time correlation into account, which results in low prediction accuracy, a temperature prediction model using long short term memory network based on random forest was proposed.The characteristic that meteorological elements are highly correlated with the temperature was selected as input variables using random forest to eliminate the noise in the original meteorological data and reduce the complexity of the network.On this basis, the long short term memory network was used to establish the overall prediction model, and the experiment was carried out on the collected multi element meteorological data.Experimental results show that the proposed model has high prediction accuracy and strong generalization ability in dealing with large scale multivariable time series data.