Research On Rapid Simulation And Prediction Method Of Urban Waterlogging Based On Machine Learning

Under the background of climate chan ge and rapid urbanization process,urban waterlogging disasters caused by extreme weather become more and more serious.Ifthere is no response means such as early warning and forecast,it will lead to serious loss of life and property.It is an important way to reduce the losses caused by urban waterlogging disasters to provide decision basis and buffer time for relevant emergency departments through rapid simulation and forecast of urban waterlogging.Therefore,how to realize the rapid and accurate simulation forecast of urban waterlogging has important research value.The traditional numerical model based on physical process can carry out fine simulation of waterlogging disaster but lacks timeliness,while the machine learning algorithm can make up for this defect by establishing relationship through data-driven.Therefore,this study aims to combine the advantages of the two and propose a rapid simulation and prediction method of urban flooding peak.Taking Haishu District of Ningbo City,Zhejiang Province as the research area,a fast forecast model was established to reduce the losses caused by rainstorm-induced waterlogging events.The main research contents and achievements are as follows:(1)A rapid prediction model of urban flood peak based on machine learning algorithm is proposed.In this model,random forest algorithm and K-nearest neighbor’algorithm are used to establish the relationship between rainfall characteristic parameters and numerical model simulation results,so as to avoid solving complex hydrodynamic equations.Meanwhile,Bayesian optimization algorithm is used to determine model parameters,and Pearson correlation coefficient is introduced to screen out parameters with poor correlation to improve accuracy.Taking Haishu district of Ningbo City as the research area,a rapid prediction model of urban flood and inundating peak was built.The results show that the mixed model has better stability than the single machine learning algorithm model,and the error of most events is controlled within 1%.At the same time,measured rainfall and waterlogging data are used to analyze the prediction effect of the model.The results show that the model is stable and reliable as a whole,and the relative errors of the predicted inundation range and inundation volume can be controlled within 10%of the measured data,and the relative errors of the numerical model can be controlled within 5%.(2)A rapid prediction model of urban flooding peak value based on machine learning algorithm was proposed.The stochastic forest algorithm and K-nearest neighbor algorithm were used to establish the relationship between rainfall characteristic parameters and numerical model simulation results,so as to avoid solving complex hydrodynamic equations.Meanwhile,Bayesian optimization algorithm was used to determine the model parameters.Pearson correlation coefficient was introduced to screen out redundant parameters to improve accuracy.Finally,an error correction matrix was constructed to reduce the accumulation of errors.Taking Haishu district of Ningbo City as the research area,a rapid prediction model of urban flood and inundating peak was built.The results show that the mixed model has better stability than the single machine learning algorithm model,and the error of most events is controlled within 1%.At the same time,measured rainfall and waterlogging data are used to analyze the prediction effect of the model.The results show that the model is stable and reliable as a whole,and the relative errors of the predicted inundation range and inundation volume can be controlled within 10%of the measured data,and the relative errors of the numerical model can be controlled within 5%.(3)To solve the problem of insufficient timeliness of the traditional numerical model,the hydrological and hydrodynamic model was constructed by using the topographic data with grid resolution of 5m as the input condition.Comparing the operation time of the numerical model based on physical process with that of the urban waterlogging rapid prediction model,the average simulation time of the rapid prediction model was 17.24s,which was 374 times faster than that of the numerical model.It can meet the requirement of fast forecast of urban waterlogging and provide sufficient preparation time for emergency decision-making of urban waterlogging.