Research On COVID-19 Prevention And Control Method Based On Intelligent Optimization

The 2019 outbreak of COVID-19 has severely affected every country/region in the world.To isolate people and slow the spread of the virus,governments have implemented suitable NPIs,including closing public places,working from home,and wearing masks.However,although these actions reduce the risk of infection,they frequently lead to a partial or entire closure of the region,which has a large financial on society.Therefore,the rational development and implementation of NPIs is a pressing issue for every country/region until an effective vaccine is developed.In this thesis,we address the problem of optimizing the combination of NPIs by using the epidemic data of each country/region since May 2020 as the research object.Firstly,this thesis analyzes the interventions that have the most significant impact on the epidemic,applying them to the epidemic prediction model for training,using an ES to improve the prediction ability of the model,and using a non-dominated ranking genetic algorithm to optimize the interventions and use the prediction model as an evaluation function to develop intervention plans to achieve a dynamic balance between reducing the number of infections and reducing the cost to society.The main research work is as follows:1)A forecasting model based on the LSTM is developed.The model fully utilizes the advantages of the LSTM model in processing time series to overcome the time lag problem and selects 12 interventions that have a significant direct impact on the epidemic’s spread are chosen as the model’s inputs.To improve the prediction ability of the model,an ES-based LSTM prediction model is proposed to optimize the hyperparameters in the neural network,and then optimize the LSTM model.2)To address the issue that the traditional NSGA-Ⅱ algorithm lacks local search capability,an improved cross-variance operator is introduced to dynamically adjust the cross-probability and variance probability by the fitness values of individuals in the population,which improves the algorithm’s search capability and convergence speed.3)The non-dominated ranking genetic algorithm based on the tabu search(TS-NSGA-Ⅱ)is proposed for developing a combination of interventions.The algorithm aims to reduce the number of new infections per day and the total cost of interventions and combines NSGA-Ⅱ with the taboo search algorithm to make up for the shortcomings of NSGA-Ⅱ in terms of local search capability.Experiments prove that the methods used in this thesis all achieve good results.The ES-LSTM model built for the epidemic prediction model has been improved in terms of prediction capability.Compared with the traditional Linear model,RNN model,and LSTM model,the algorithm improved the MAE by 72.9%,27.6%,and 26.3%,and the RMSE by 74.15%,31.4%,and 29.5%,respectively;the improved NSGA-Ⅱ algorithm showed better convergence and distribution in the standard test function,ensuring the population of stable diversity;the TS-NSGA-Ⅱ model established for the optimization of NPIs can obtain a better set of Pareto frontier solutions,compared with the greedy algorithm,random assignment algorithm,NEAT model and NSGA-Ⅱ algorithm.