Data-Driven Fire Risk Prediction And Rescue Difficulty Assessment

With the accelerating of urbanization process,construction units are increasingly getting denser,and correspondingly the urban fire rescue work has become more severe.It is too tough to efficiently complete the tasks such as fire inspection,fire rescue and so on,under limited manpower for the fire department.Therefore,reducing the workload without reducing the effectiveness of the fire rescue has become one of the important directions in the current work of the fire department.The fire prevention work of urban buildings is attracting much attention from all walks of life.In this thesis,work is divided into three aspects: fire inspection,fire warning and fire rescue.By analyzing the historical fire data accumulated by fire departments,we construct a model to optimize the three problems above with machine learning algorithm.The aim of the research is to provide decision support for the fire departments.This thesis mainly includes the following three parts: relative factor analysis of fire,unit fire risk prediction model and rescue difficulty evaluation model.In the relative factor analysis of fire,this thesis uses data statistics to initially screen firerelated factors,then to further determine the impact of certain factors on fire by using correlation analysis.Finally,the data set is constructed and frequent item set mining algorithm is used to mine the frequent occurrences of attribute combination in the fire data sets.In the unit fire risk prediction research,this thesis uses the Extreme Gradient Boosting algorithm(XGBoost)combined with historical fire-related data to construct a unit fire risk prediction model to achieve an advance prediction of fire risk of the unit in the future time period,and conducts a targeted inspection on high-risk units.The fire risk prediction model construction mainly includes four parts: data preprocessing,training set construction,model training and model evaluation.The experimental result shows that it is feasible to predict the unit fire risk by using the XGBoost algorithm.In the research of fire rescue difficulty evaluation model,this thesis proposes a multiattribute decision-making method based on multiple factors to assess the fire rescue difficulty,and constructs the fire rescue difficulty prediction model with XGBoost algorithm so as to realize the quantification and advance prediction of the rescue difficulty of the unit fire,whereby,help the fire department accurately dispatch firefighting brigade.