基于经验模态分解的CEEMDAN-FE-LSTM传染病预测

Prediction of Infectious Diseases Using CEEMDAN-FE-LSTM Model Based on Empirical Mode Decomposition

针对现有的传染病预测模型未充分考虑到时间序列的复杂度,提出了一种基于自适应噪声完备集合经验模态分解(CEEMDAN)和模糊熵(FE)改进长短时记忆网络(LSTM)的传染病组合预测模型.首先,运用CEEMDAN算法将序列分解成若干个不同频率的模态分量与残差分量,以降低原始时间序列的复杂度;然后,运用FE算法计算各分量的时间复杂度,并将其重构为不同尺度的序列以提高运算效率;最后,建立LSTM模型对重构序列分别进行预测,得到最终预测结果.根据2010年1月至2021年12月肺结核、乙肝、布鲁氏菌病和艾滋病发病数据进行模型预测,并与SARIMA模型、CEEMDAN-FE-SARIMA模型和LSTM模型进行对比.结果表明,提出的模型较常规模型可以更好地把握传染病发病的变化规律,降低时间序列的复杂度,提高传染病预测精度.

To address the problem that the existing infectious disease prediction models do not fully consider the complexity of time series,an improved combined infectious disease prediction model based on Complete Ensemble Empirical Mode Decomposition with Adaptive Noise(CEEMDAN)and Fuzzy Entropy(FE)for Long and Short-Term Memory(LSTM)is proposed.First,the CEEMDAN algorithm is applied to decompose the sequence into several modal components with different frequencies and residual components.Then,the FE algorithm is applied to calculate the time complexity of each component and reconstruct them into sequences of different scales.Finally,an LSTM model is built to predict the reconstructed sequences separately to obtain the final prediction.The data of the number of incidence of tuberculosis,hepatitis B,brucellosis and AIDS from January 2010 to December 2021 are selected and compared with the SARIMA model,CEEMDAN-FE-SARIMA model and LSTM model in this paper.The experimental results show that the model can better grasp the change pattern of infectious disease incidence and can effectively improve the accuracy of infectious disease prediction.