基于网络信息与BP神经网络的应急物资需求预测研究--以积石山县6.2级地震为例

Emergency Material Demand Prediction Based on Network Information and BP Neural Network:Taking Jishishan M S6.2 Earthquake,Gansu in 2023 as an Example

为应对震后应急初期难以及时获取灾情信息,灾区内各区域应急物资需求不明等问题,引入实时网络资源信息结合历史地震数据进行物资需求预测。通过爬取2023年12月18日甘肃积石山地震震后72h内112672条多平台网络信息,使用BERT-CNN模型对网络信息进行分类。结合历史地震数据,采取TOPSIS方法完成应急需求急迫度评估,将其作为新的样本特征引入BP神经网络,以优化对积石山县内各乡镇死亡人数预测效果。最后基于死亡人数与安全库存理论,完成对震后72h灾区各乡镇饮用水、帐篷两类物资的需求预测。通过实验证明,引入网络资源可以及时反映灾区受灾情况,提高应急物资预测的时效性、精细度和准确性。

To deal with the difficulties of obtaining timely disaster information in the early stages of post-earthquake emergency response,as well as the unknown demand for emergency supplies in different areas of the affected area,we introduced the real-time network resource information for predicting material needs combined with historical earthquake data.By crawling 112,672 multi-platform network information within 72 hours after the Jiashishan M S6.2 earthquake in Gansu on December 18,2023,the network information is classified with the use of the BERT-CNN model.Combining with the historical earthquake data,the urgency of emergency needs is assessed by the TOPSIS method,which is then used as a new sample feature into the BP neural network to optimize the prediction of death toll in various towns and villages in Jishishan County.Finally,based on the death toll and the theory of safety stock,the demand for drinking water and tents in each township of the disaster-stricken area within 72 hours after the earthquake is predicted.As demonstrated by the experiments,the introduction of network resources can promptly reflect the disaster situation in the earthquake affected area,and could improve the timeliness,precision,and accuracy of emergency material prediction.