Abstract
Instrumental seismic intensity acts as a quantitative measure of seismic impacts, as detected by monitoring instruments. In this study, an effective and accurate model has been developed to evaluate this intensity, using a dataset consisting of 159 seismic records collected from western China. A gray correlation analysis technique is applied to establish relationships between different ground motion parameters and macro-intensity. The parameters with strong correlation coefficients are determined as the input of the machine learning model, including Clough SI, peak ground acceleration, root-mean-square acceleration, effective peak acceleration, effective peak velocity, Housner SI, peak acceleration response spectrum, peak velocity response spectrum, and cumulative absolute velocity, and the macro intensity is specified as the expected output.By leveraging machine learning algorithms, specifically the multilayer perceptron (MLP) and random forest (RF) within the TensorFlow framework, this model illuminates the relationships among the input parameters. The efficacy of the model is assessed using accuracy, mean square error (MSE), and mean absolute percentage error (MAPE). Subsequent validation with a test set emphasizes the robustness of the model, particularly highlighting the strong correlation between Clough SI and PGA with macro survey intensity. Comparative analysis with traditional methods indicates that the proposed machine learning algorithms improve accuracy rates, with RF and MLP models achieving accuracies of 79.17% and 72.92%. In particular, the RF model exhibits superior performance in terms of accuracy, MSE, and MAPE, demonstrating its higher stability and precision in assessing seismic intensity.