基于小波变换和LSTM混合模型的气象时间序列预测研究

Research on Meteorological Time Series Prediction Based on Wavelet Transform and LSTM Hybrid Model

摘要: 随着生态环境意识的提高和科技的进步，气象数据的预测研究逐渐成为深度学习领域的热门话题。气象数据不仅是气象领域研究中极其重要的资料，也可应用于环境、能源等诸多领域的科学研究，对我国经济和生态环境发展起着重要作用。本文针对气象数据多波动、非线性且存在极端数据等特点，构建了经小波变换降噪处理后基于LSTM神经网络的混合预测模型。该模型首先通过小波变换将原始气象数据分解为低频和高频分量后再重新组合去噪，继而使用LSTM网络模型建模预测，并以未来时刻的气象指标作为最终预测结果。本文以平均风速、地表温度和降水为例进行实证验证。研究结果表明，与长短期记忆神经网络(Long Short-Term Memory, LSTM)等单一模型相比，本文提出的预测方法具有更高的预测精度，更小的预测误差，可为气象预警、自然灾害管控等部门提供具有参考价值的方案。

Abstract: With the improvement of ecological environment awareness and the progress of science and technology, the prediction research of meteorological data has gradually become a hot topic in the field of deep learning. Meteorological data are not only extremely important data in meteorological research, but also can be used in scientific research in many fields such as environment and energy, and play an important role in the development of economy and ecological environment in China. Due to the characteristics of meteorological data with multiple fluctuations, nonlinearity and extreme data, this paper constructs a hybrid prediction model based on LSTM neural network after wavelet transform and noise reduction. The model first decomposes the original meteorological data into low-frequency and high-frequency components through wavelet transform, and then denoises and recombines them. Then, the LSTM network model is used to model and predict, and the meteorological indicators in the future are used as the final prediction result. This paper takes the average wind speed, surface temperature and precipitation as the final prediction goal. The research results show that, compared with a single model such as Long Short-Term Memory (LSTM) neural network, the prediction method proposed in this paper has higher prediction accuracy and smaller prediction error, which can provide solutions with reference value for meteorological early warning, natural disaster management and other departments.