School of Mathematics Physics and Statistics, Shanghai University of Engineering Science, Shanghai 201620, China.
Comput Math Methods Med. 2022 Apr 4;2022:1566727. doi: 10.1155/2022/1566727. eCollection 2022.
Since the outbreak of COVID-19, BRICS countries have experienced different epidemic spread due to different health conditions, social isolation measures, vaccination rates, and other factors. A descriptive analysis is conducted for the spread of the epidemic in the BRICS countries. Considering the nonlinear and nonstationary characteristics of COVID-19 data, a principle of decomposition-reconstruction(R)-prediction-integration is proposed. Correspondingly, this paper constructs an integrated deep learning prediction model of CEEMDAN-R-ILSTM-Elman. Specifically, the prediction model is integrated by complete ensemble empirical mode decomposition (CEEMDAN), improved long-term and short-term memory network (ILSTM), and Elman neural network. First, the data is decomposed by adopting CEEMDAN. Then, by calculating the permutation entropy and average period, the decomposed eigenmode component IMFs are reconstructed into four sequences of high, medium, low level, and trend term. Thus, ILSTM and Elman algorithms are used for component sequence prediction, whose results are integrated as the final results. The ILSTM is established based on the LSTM model and the improved beetle antennae search algorithm (IBAS). The ILSTM mainly considers that the prediction accuracy of LSTM model is vulnerable to the influence of parameter selection. The IBAS with adaptive step size is used to automatically optimize the super parameters of LSTM model and to improve the modeling efficiency and prediction accuracy. Experimental results indicate that compared with other benchmark models, CEEMDAN-R-ILSTM-Elman integrated model predicts the number of newly confirmed cases of COVID-19 in BRICS countries with higher accuracy and lower error. Strict social policies have a greater impact on the infection rate and mortality rate of the epidemic. During July-August 2021, epidemic spread in BRICS countries will slow down, and the overall situation is still quite severe.
自 COVID-19 爆发以来,金砖国家由于健康状况、社会隔离措施、疫苗接种率等因素的不同,经历了不同的疫情传播。对金砖国家的疫情传播进行描述性分析。考虑到 COVID-19 数据的非线性和非平稳性特征,提出了分解-重构(R)-预测-集成的原理。相应地,本文构建了一个集成深度学习预测模型的 CEEMDAN-R-ILSTM-Elman。具体来说,该预测模型由完全集合经验模态分解(CEEMDAN)、改进的长短期记忆网络(ILSTM)和 Elman 神经网络集成而成。首先,采用 CEEMDAN 对数据进行分解。然后,通过计算排列熵和平均周期,将分解得到的本征模态分量 IMF 重构为高、中、低水平和趋势项四个序列。因此,采用 ILSTM 和 Elman 算法对分量序列进行预测,其结果作为最终结果进行集成。ILSTM 是基于 LSTM 模型和改进的甲壳虫触角搜索算法(IBAS)建立的。ILSTM 主要考虑到 LSTM 模型的预测精度容易受到参数选择的影响。采用具有自适应步长的 IBAS 自动优化 LSTM 模型的超参数,提高建模效率和预测精度。实验结果表明,与其他基准模型相比,CEEMDAN-R-ILSTM-Elman 集成模型对金砖国家 COVID-19 新确诊病例数的预测具有更高的准确性和更低的误差。严格的社会政策对疫情的感染率和死亡率影响更大。2021 年 7-8 月,金砖国家疫情传播速度将放缓,整体形势仍相当严峻。