To predict the distribution of instrumental seismic intensities around the epicenter immediately after an earthquake, we selected 2 331 triggered strong motion stations from 632 earthquakes and extracted the epicentral distance, magnitude, and seven effective ground motion parameters within 20 seconds after the triggers to the stations. The selected data samples were trained using an artificial neural network, and three effective prediction models were established. The results show the following:(1)Model one, which selects seven ground motion parameters without epicentral distance and magnitude, has good timeliness in prediction. From 1 to 20 s, the average intensity difference in prediction gradually reduces to 0.45.(2)Model two selects eight ground motion parameters, including the epicentral distance; it can be used to ensure early earthquake warning through the prediction of magnitude. The average intensity difference in prediction gradually reduces to 0.36.(3)Model three selects nine ground motion parameters and has the best prediction result among the three models; its average intensity difference in prediction reduces to 0.31, and it can be used to predict the post earthquake intensity field in real time. The three models were used to predict the intensity of two real earthquakes, and the difference in predicted intensity is greater than 95% and 76%, within 1, indicating that the models can be applied in earthquake early warning.
李水龙,陈以伦,于伟恒,周施文.基于人工神经网络的仪器地震烈度预测模型研究[J].地震工程学报,2023,(1):181-190. LI Shuilong, CHEN Yilun, YU Weiheng, ZHOU Shiwen. Prediction models for instrumental seismic intensity based on artificial neural network[J]. China Earthquake Engineering Journal,2023,(1):181-190.复制