基于网络熵探测新冠爆发临界信号

Detection of Critical Signals in Local COVID-19 Outbreaks Based on Landscape Network Entropy

2019年冠状病毒疾病(COVID-19)的迅速传播对世界各地的人们构成了巨大威胁,必须制定有效的策略来检测COVID-19爆发的预警信号以便及时采取适当的控制措施。与时间序列预测不同,疫情爆发通常是非线性的事件,其特征是从缓慢变化到急剧转变,因此难以预测。通过大量采集地理区域网络及其日增确诊病例的实时数据的动态信息,本文采用了一种非线性的无模型方法,即网络熵(LNE)方法,以识别检测出新冠疫情进行灾难性转变之前的前爆发阶段。在对包括中国湖北省、日本关东地区以及巴西部分地区在内的多个地区的新冠疫情爆发临界点的预测上,网络熵方法都取得了非常理想的预测效果。

The rapid spread of coronavirus disease 2019 (COVID-19) has posed an enormous threat to people all around the world. It is imperative to develop effective strategies for detecting early-warning signals of COVID-19 outbreaks to implement appropriate control measures in a timely manner. Unlike time-series prediction, outbreaks are generally highly nonlinear events with characteristics that develop from gradual changes to drastic transitions, and are thus difficult to predict. By exploiting dynamic information collectively from geographic district networks and the real-time data of daily new cases, we adopted a nonlinear model-free approach, the landscape network entropy (LNE) method, to identify the pre-outbreak stage prior to the catastrophic transition into a COVID-19outbreak. The LNE method was successfully validated by accurate predictions of the local COVID-19 outbreaks in various regions, including Hubei Province of China, the Kanto region of Japan and parts of Brazil.