Data-driven Prediction Of The Evolution Of Unexpected Epidemic Outbreaks

As the most serious unexpected epidemic outbreaks in the world in recent years,COVID-19 has threatened the lives of people around the world and caused serious damage to global economic security.Therefore,timely prediction of the evolution of the epidemic is crucial for epidemic prevention and control.At present,the epidemic disease model based on system dynamics is a relatively mature and commonly used unexpected epidemic outbreaks prediction model,and data-driven machine learning also provides a new effective way for the evolution prediction of unexpected epidemic outbreaks.Based on the summary of relevant studies at home and abroad,this paper conducts in-depth research on the application of traditional SEIR model and LSTM model in unexpected epidemic outbreaks prediction,and attempts to improve some aspects of these models,as follows:1)When using traditional SEIR models to predict the evolution of the epidemic,the accuracy of the prediction tends to decline over time.In view of this problem,this paper takes COVID-19 in Wuhan as the background,combines the evolutionary characteristics of the epidemic and the government’s epidemic prevention measures,proposes a four-stage SEIR model considering actual interventions.It also analyzes the impact of various emergency measures on the effectiveness of epidemic prevention and control,and estimates the basic reproduction number of COVID-19 in Wuhan.The test results show that the four-stage SEIR model designed in this paper can effectively predict the peak,scale and duration of the spread of COVID-19 in Wuhan.At the same time,this paper uses Baidu migration index to help identify the whereabouts of potential patients,so as to give an early warning of the main population outflow target areas before the closure of Wuhan.2)When applying the classic LSTM model to predict the evolution of the epidemic,the hyperparameters are usually manually adjusted,but the commissioning work is huge and timeconsuming.In view of this problem,this paper uses sparrow search algorithm(SSA)for LSTM model hyperparameter optimization,so as to realize the data-driven epidemic evolution prediction.In order to verify the prediction performance of the above combined method,the cumulative number of COVID-19 infections in Brazil,India and Colombia was taken as the prediction target.Meanwhile,loss function was used to evaluate the results.Finally,cuckoo search algorithm(CS)and particle swarm optimization algorithm(PSO)were used for comparative analysis.The test results show that SSA can effectively improve the accuracy of LSTM model,and the SSA-LSTM combined method proposed in this paper has better performance in the evolution prediction of COVID-19.3)Since the above two prediction methods have their own advantages and disadvantages,this paper further attempts to combine system dynamics models and machine learning methods to construct SSA-LSTM-SIRD models,and carries out modeling and analysis from the perspective of series and parallel.Similarly,this paper uses a combined model constructed to predict confirmed and removed cases of COVID-19 in Brazil,India,and Colombia.The test results show that the series model is superior to the parallel model,and the two combined models show advantages in predicting the epidemic evolution of the three countries,with good prediction accuracy and stability.