Rolling Prediction of Emergency Supplies Based on Postdisaster Multisource Time-Varying Information

Accurate prediction of material demands is key to ensuring the overall efficiency of emergency rescue operations. From the perspective of the prediction method, the single-material static prediction method based on the overall data has limitations. This method cannot flexibly adjust multiperiod material demands. Considering data sources, acquiring data regarding material demands in historical disasters is more difficult and has more uncertainty compared with statistical data on deaths. This study investigates a rolling prediction method for emergency supplies based on postdisaster multisource time-varying information to ensure prediction accuracy. First, the proposed method uses historical cases, real-time disasters, and time-sharing simulation data as the source data. The method implements attribute reduction of original data samples based on rough set theory and predicts cumulative death tolls in each rolling period by using the rolling time-domain as the basic framework and combining a support vector machine (SVM). Second, the proposed method estimates the material demands in the corresponding period by using the material demand model according to the prediction results in a single period. Finally, the proposed method is verified by an experiment with a general mean prediction error of 10.96%. However, the general mean prediction error of SVM reaches 17.77% in the static multistep prediction. Moreover, the general mean prediction error of the methods in the references is 14.13%. Overall, the method has high accuracy and strong timeliness. Prediction results can not only be used as a basis for material estimation, but also provide several scientific and effective references for the allocation and scheduling of emergency supplies.