To rapidly and accurately provide alerts at target sites near the epicentre, we develop an on-site alert-level earthquake early warning (EEW) strategy involving P-wave signals and machine-learning-based prediction equations. These prediction equations are established for magnitude estimation and peak ground velocity (PGV) prediction accounting for multiple feature inputs and the support vector machine (SVM). These prediction equations are called SVM-M model for estimating magnitude and SVM-PGV model for predicting PGV, respectively. According to comparison between the predicted magnitude and PGV values with the predicted threshold values (M = 5.7 and PGV = 9.12 cm s^–1, respectively), different alert level (0, 1, 2, 3) is issued at the different recording site when the predicted magnitude or PGV values exceed the given threshold values. Alert level 3 means that both the predicted magnitude and the predicted PGV exceed a given threshold, and there may be serious damage in this recording site. We apply the method to three destructive earthquake events (M ≥ 6.5) occurred in Japan, and our results indicate that with regard to the performance of SVM-PGV model for predicting PGV, at 3 s after P-wave arrival, the percentage of successful alarms (SAs) for these three events is higher than 95, 73 and 94 per cent, respectively, and the percentage of false alarms approaches 0. Additionally, with regard to the performance of SVM-M model for estimating magnitude, at 3 s after P-wave arrival, the percentage of SAs for these three events exceeds 95 per cent, and the percentage of missed alarms approaches 0. Moreover, almost all stations in the areas PGV ≥ 16 cm s^–1 (I_MM ≥ VII) near the epicentre issue alert level 3. The proposed method provides potential applications in EEW system.

In this paper, a nonlinear regression method called a support vector regression (SVR) is presented to establish the relationship between engineering ground motion parameters and macroseismic intensity (MSI). Sixteen ground motion parameters, including peak ground acceleration (PGA), peak ground velocity (PGV), Arias intensity, Housner intensity, acceleration spectrum intensity, velocity spectrum intensity, and others, are considered as candidates for feature selection to generate optimal SVR models. The datasets with both useable strong ground motion records and corresponding investigated MSIs in the Sichuan–Yunnan region, China, are all collected, and these 125 pairs of datasets are used for selecting features and comparing regression results. Nine ground motion parameters are selected as the most relevant features: PGA is the first fundamental one and PGV is the fifth relevant feature. Based on performance measures on the testing dataset, the best SVR model is given when the number of features is one all the way up to nine. According to predicted accuracy, SVR models with Gaussian kernel give much better MSI prediction than linear kernel SVR models and linear regression models. In particular, the Gaussian kernel SVR of PGA gives much higher MSI prediction accuracy than the linear regression model of PGV and PGA. The proposed SVR models are valid for MSI values from VI to IX, and they can be used for rapid mapping damage potential and reporting seismic intensity for this high-seismic-activity region.

The Sichuan–Yunnan region is a seismically active area. To explore the feasibility of using the support vector machine (SVM) method for magnitude estimation in the area and to improve the rapid magnitude estimation accuracy, we construct an SVM magnitude estimation model using transfer learning (TLSVM‐M model) based on a single‐station record in this study. We find that the magnitude estimation of a single station shows that for the test dataset, within the 3 s time window after the P‐wave arrival, the average absolute error (which reflects the size of the estimated magnitude error as a whole) and standard deviation (which reflects the scatter of magnitude estimation error) of the magnitudes estimated by the TLSVM‐M model are 0.31 and 0.41, respectively, which are less than those of the SVM magnitude estimation model without transfer learning (0.44 and 0.55, respectively), the $τc$ method (1.35 and 1.74, respectively) and the $Pd$ method (0.44 and 0.56, respectively). In addition, in test involving five earthquake events via the TLSVM‐M model, at 1 s after the first station is triggered, the magnitudes of three events (⁠ $Ms$ 4.2, 5.2, and 6.3) are estimated within an error range of ±0.3 magnitude units. For the other two earthquakes (⁠ $Ms$ 6.6 and 7.0), there is an obvious magnitude underestimation problem at 1 s after the first station is triggered, with less underestimation by increasing time after the first station is triggered. Meanwhile, for these two events (⁠ $Ms$ 6.6 and 7.0), within 13 s after the first station was triggered, the magnitude estimation errors are both within ±0.3 magnitude units. The TLSVM‐M model has the capability of rapid magnitude estimation for small‐to‐moderate events in the Sichuan–Yunnan region. Meanwhile, we infer that the proposed model may have potential in earthquake early warning.

The traditional magnitude estimation method, which establishes a linear relationship between a single warning parameter and the magnitude, exhibits considerable scatter and underestimation. In addi...

The accurate and reliable discrimination of earthquakes from background noise is a primary task of earthquake early warning (EEW); however, ubiquitous and complex microtremor signals substantially complicate this task. To mitigate this problem, a generative adversarial network (GAN) is adopted to distinguish between earthquakes and microtremors in this study. We train a GAN based on 52,537 K‐NET and KiK‐net strong ground motion records from Japan, and use the well‐trained discriminator to identify 5373 P waves and 5373 microtremors in the testing set. The results indicate that this algorithm can correctly identify 99.89% of P waves and 99.24% of microtremors with high confidence. In addition, a verification of the proposed algorithm on data from the Great East Japan earthquake confirms that this model can achieve robust results for local records of large events and ultimately discriminate earthquakes from microtremors. This algorithm is an exploratory test of a GAN for identifying earthquake P waves. Though the GAN uses only P waves for training (There are no microtremors in the input data.), it has extensive potential in seismological and EEW applications.

Earthquake early warning (EEW) is of great significance in mitigating seismic disasters. Traditional EEW algorithms, which are knowledge‐driven approaches, rely on seismologists' analysis. The limited intensity measures were extracted by seismologists from P‐wave signals. And there is considerable uncertainty for predicting epicentral distance, magnitude, peak ground acceleration (PGA), and peak ground velocity (PGV). Currently, data‐driven deep learning methods with the strong learning abilities do not consider knowledge information from seismologists in EEW; thus, there is unexplored potential in enhancing the performance of deep learning models for EEW. Here, we construct the Data‐knowledge driven Hybrid deep Learning network (DHLnet) for EEW using the waveform input, knowledge embedding, convolutional neural network and graph convolutional network, aiming to integrate knowledge information from knowledge‐driven methods and the strong learning ability of data‐driven deep learning methods, that is, improving the performance of EEW. For the same test data set, compared with knowledge‐driven methods and data‐driven deep learning models, we demonstrate that DHLnet enhances the timeliness and robustness in predicting the epicentral distance, magnitude, PGA, and PGV during 10 s time window following the arrival of P‐wave. Furthermore, to validate the generalization and robustness of the DHLnet in EEW, we applied the trained DHLnet to an independent data set, within first few seconds after an earthquake occurs, DHLnet can provide robust magnitude estimation, epicentral distance estimation and high alarm accuracy. The potential of the proposed network is to enhance the performance of EEW systems and provides new insights into the exploration of deep learning methods for EEW domain.