Abstract:

Since the outbreak of Corona Virus Disease 2019（COVID-19） in Wuhan, China at the end of December 2019, my country’s economy and society have suffered great harm. Using network data to warn the development trend of the epidemic can effectively reduce its social harm. When using machine learning algorithms to build an early warning model, parameter selection is an important content, which is closely related to the accuracy of the final built model. It explores the application effects of a variety of new intelligent optimization algorithms in the Baidu search index COVID-19 early warning model, which can provide a certain theoretical basis and analysis strategy for the popularization and application of new intelligent optimization algorithms. Three new intelligent optimization algorithms, Multi-Verse Optimizer（MVO）, Slime Mould Algorithm（SMA） and Equilibrium Optimizer（EO）, are compared in the application effect of the Least Squares Support Vector Machine（LSSVM） in Baidu search index epidemic warning model. It optimizes algorithm optimization process, the SMA algorithm has poor convergence, and the global search ability is weaker than the MVO and EO algorithms, while the EO algorithm has relatively low computational efficiency. The MVO algorithm has high computational efficiency and strong convergence. Finally, an early warning model is constructed. The advantage is obvious（MSE=17.77, MAE=38.38, RMSE=129.3, R2=0.87）. The three intelligent optimization algorithms can all improve the prediction performance of the LSSVM early warning model, but the MVO optimization algorithm has the best comprehensive computing performance. The final MVO-LSSVM early warning model can be used to predict epidemic prevention behaviors in the subsequent epidemic prevention and control stage certain reference.

Key words: intelligent optimization algorithm, Baidu index, early warning model, comparative study, COVID-19

摘要：

自2019年12月底中国武汉爆发新型冠状病毒性肺炎（Corona Virus Disease 2019，COVID-19）以来，我国经济社会遭受巨大危害，利用网络数据预警疫情发展趋势可以有效降低其社会危害。而采用机器学习算法构建预警模型时，参数选取是其中重要内容，与最终构建模型的精度密切相关，探讨多种新型智能优化算法在百度搜索指数COVID-19预警模型中的应用效果，可为新型智能优化算法的推广应用提供一定的理论依据和分析策略。对比多元宇宙算法（Multi-Verse Optimizer，MVO）、黏菌算法（Slime Mould Algorithm，SMA）及平衡算法（Equilibrium Optimizer，EO）三种新型智能优化算法，在最小二乘支持向量机（Least Squares Support Veotor Machine，LSSVM）百度搜索指数疫情预警模型中的应用效果。优化算法寻优过程，SMA算法收敛性较差，全局搜索能力弱于MVO和EO算法，而EO算法运算效率相对较低，MVO算法的运算效率高，收敛性也较强，最终构建预警模型优势明显（测试集的MSE为17.77，MAE为38.38，RMSE为129.35，R2为0.87）。三种智能优化算法皆可提升LSSVM预警模型的预测性能，而MVO优化算法的综合运算效能最好，最终构建的MVO-LSSVM预警模型可为后续疫情常态化防控阶段的防疫行为预判提供一定参考。

关键词: 智能优化算法, 百度指数, 预警模型, 对比研究, 新冠疫情