An Integrated Learning-Based Seismic Damage Assessment Method for Single Buildings

Yatian Meng
Department of Surveying and Geo-Informatics, Southwest Jiaotong University, Chengdu 611756, Sichuan Province, China
Yongliang Xiong
Department of Surveying and Geo-Informatics, Southwest Jiaotong University, Chengdu 611756, Sichuan Province, China
Hongmei Guo
Sichuan Earthquake Agency, Chengdu 610041, Sichuan Province, China
Ying Zhang
Sichuan Earthquake Agency, Chengdu 610041, Sichuan Province, China

Abstract:

Seismic damage assessment of buildings is of great significance for pre- and post-earthquake emergency management. In recent years, integrated learning methods, as an excellent machine learning method with high accuracy and strong generalization ability, have been widely used in various research fields. In this paper, three integration learning methods, namely Random Forest algorithm, XGBoost algorithm and LightGBM, are investigated and a single building assessment method based on integration learning is proposed. On the basis of the historical earthquake case data and building survey data of Sichuan Province, a sample database of the model is established and a combined SMOTE-Tomek sampling technique is used to correct the effects of category imbalance in the original dataset. Then, three integrated learning methods were used to train the evaluation models respectively, and the evaluation results of different methods were obtained. The experiments compare the performance of Random Forest algorithm, XGBoost algorithm and LightGBM algorithm on the test dataset, and the results show that the evaluation accuracy of all three algorithms is above 80%, among which the accuracy of the model based on XGBoost algorithm is the highest with 87.26%, and the accuracy of the model based on Light GBM algorithm is higher with 85.35% with the fastest arithmetic speed. The results show that the single building assessment method based on integrated learning proposed in this paper can assess the damage of buildings more accurately.