Research Of Seismic Casualties Assessment Model Based On Machine Learning Algorithm

Many earthquakes occur every year around the world,accompanied by varying degrees of casualties and economic losses.The number of casualties is quickly obtained after the earthquake,and the results combined with expert experience provided scientific guidance for emergency rescue work after the earthquake.This can make the distribution of earthquake victims and materials more reasonable,and reduce unnecessary economic losses and casualties caused by untimely rescue effectively.This article first selects the data for the assessment of earthquake losses in mainland China from 1949 to 2017.Then,the appropriate influencing factors of earthquake casualties and suitable earthquakes are selected based on the machine learning algorithm.Finally,a post-earthquake casualty quick assessment model is established.The main content of this article are as follows:1.The correlation and research progress of the influencing factors of earthquake casualties are analyzed.Several types of earthquake casualty assessment methods commonly,and their applicable conditions and background are summarized.The research status of the cross-algorithm of machine learning and seismology is introduced.Combining the complexity of the factors,the defects of the previous prediction methods and the applicability of machine learning in the field of earthquake engineering,the method used to evaluate the casualties of earthquake is determined.2.Through the classification and basic principles of Stacking,Bagging and Boosting algorithms in the ensemble learning,it focuses on the steps of analysis of the feature importance of basic learner CART decision tree,Ada Boost of the Boosting family and random forest algorithm of the Bagging family.The three algorithms determined to select suitable factors of earthquake casualties,and the most suitable algorithm is selected for subsequent calculation.Then,the artificial neural network frame,deep generation model,training and optimization methods and framework in the deep learning algorithm are analyzed to determine the framework.3.The data of the destructive earthquakes that occurred between 1949 and 2017 are selected.An appropriate earthquake case set is selected and the default value is processed based on the data preprocessing method and earthquake experience.Three ensemble learning algorithms are used to establish the evaluation model of the importance of factors.The preliminary calculation results are analyzed and the factors are used for further screened.Once again,an evaluation model is established to sort and analyze the screened factors.The accuracy and stability of the algorithms are compared,and the optimal algorithm is selected for the next calculation.Then,the structural types are classified,and the random forest algorithm is used to evaluate the contribution of different types to the number of earthquake deaths.The ranking of the contribution of the obtained types is analyzed.According to the results,the seismic performance of the structure is analyzed and reasonable building construction suggestions are given.The house destruction influence coefficient is given based on the contribution of the house type,and appropriate data is selected based on the integrity of the house data.The evaluation and screening of the influencing factors of casualties will lay a solid foundation for the subsequent house destruction influence coefficient model.4.The factors with higher importance are selected as the input layer,and the number of deaths and injuries are used as the output layer to establish a suitable deep learning model.Based on the particularity of the data,suitable hyper parameters,framework and optimization algorithms are used.The model is trained and optimized several times.Then,the analysis and comparison of the results are conducted.Some hyper parameters and input parameters are modified and a model based on the house destruction influence coefficient is built.The two models are compared,and the secondary disaster correction formula is given based on the latter result to improve the accuracy and applicability of the model.