基于支持向量机预测模型的高速铁路现地地震预警方法

On-Site Earthquake Early Warning Method for High-Speed Railway Based on Support Vector Machine Prediction Models

以准确发出预测高速铁路Ⅰ级地震警报为目标,提出基于支持向量机(SVM)预测模型的高速铁路现地地震预警方法,并分析该方法的可行性。首先,针对震级和地震动加速度峰值2个参数,分别明确相应的SVM预测模型;然后,设置阈值(震级6级和地震动加速度峰值40 cm·s-2)并定义预测等级及其与阈值间的对应关系,依据2个参数预测值与阈值的关系发布不同等级警报;最后,根据地震动加速度峰值观测值与预测值的对比,定义报警动作,评价报警性能。依托2022年1月8日青海门源6.9级地震的全部加速度数据进行离线验证,结果表明:P波到达后1~3 s时间窗内,2个参数的预测值逐渐接近真实值,预测误差的标准差明显减小,SVM预测模型可得到准确预测结果;首台(距离震中最近且离破坏最近的台站)触发后1 s时即可成功发布警报,触发后6 s时烈度Ⅶ度区域内的所有台站均成功发布警报。该方法可为低成本烈度仪服务高速铁路地震监测预警提供参考。

Aiming at accurately raising the prediction of first-level earthquake alarm for high-speed railway, this paper proposes the on-site earthquake early warning(EEW) method for high-speed railway based on support vector machine(SVM) prediction models, and analyzes the feasibility of this method. Firstly, for the two parameters of magnitude(M) and seismic peak ground acceleration(PGA), the corresponding SVM prediction models are established accordingly. Then, this paper sets the threshold values(magnitude is M6 and seismic peak ground acceleration is 40 cm·s-2), defines the corresponding relationship between the prediction levels and its threshold values, and issues different levels of alarm according to the relationship between the predicted values and the threshold values of the two parameters. Finally, according to the comparison between the observed value and the predicted value of PGA, the alarm actions are defined to evaluate the alarm performance. Based on the off-line verification of the full acceleration data of the M6. 9 earthquake in Menyuan,Qinghai Province on January 8, 2022, the results show that within a 1-3 s time window after the arrival of P wave, the predicted values of the two parameters gradually approach the actual values, the standard deviation of prediction error is significantly reduced, indicating that the SVM prediction models can obtain accurate prediction results. At 1 s after the first station(closest to the epicenter and the damage) is triggered, the alarms are successfully issued. At 6 s after the first station is triggered, all stations in the intensity VII area can successfully issue alarms. This method can provide a reference for low-cost intensity instruments to serve earthquake monitoring and early warning for high-speed railway.