Research On Analysis And Prediction Of COVID-19 Propagation Dynamics Based On Reservoir Computing

The spread of COVID-19 has led to a serious public health crisis,posing a major threat to the safety of human life and property worldwide.Therefore,forecasting the development trend of the COVID-19 epidemic is of great significance for the government and individuals to effectively carry out epidemic prevention and control.In terms of prediction methods,traditional dynamic models mainly use parameter fitting methods for prediction.However,the paradigm of conventional models is fixed,which can easily overlook the influence of potential factors,resulting in poor prediction performance.Therefore,this article first conducts targeted modeling on the prediction target and studies the prediction effect of infectious disease models.On the other hand,Reservoir Computing(RC),as a type of neural network,is widely used in time series prediction due to its simple structure and fast convergence speed.However,RC has not been well applied to the prediction of COVID-19 so far.Therefore,this paper will continue to rely on RC to carry out the prediction research on the transmission dynamics of COVID-19 epidemic,aiming to find high-precision and efficient prediction means.Based on the above two methods,this research work is mainly reflected in the following three aspects:Firstly,an innovative SEIRDQF model based on the classical infectious disease dynamics model is proposed to model and predict the mid to late stage development of the epidemic in Wuhan.In this part of the work,factors such as human control measures,medical isolation measures,and the death toll are specifically considered,and a set of time-varying isolation parameters and self isolated populations are added to the effectiveness of isolation measures.Finally,the optimal solution of the model parameters is obtained through genetic algorithm.The simulation results show that the model proposed in this paper can effectively predict the development trend of different populations in the middle and late stages of the COVID-19 epidemic in Wuhan.Secondly,in order to address the shortcomings of dynamic models that rely too heavily on previous development data and improve prediction accuracy,a time-varying weight model based on RC is proposed.In this part of the work,a unique synchronous prediction mode is proposed based on the structure of RC itself and the characteristics of the processed data.Secondly,the optimal hyperparameter of RC in the prediction mode is deduced by the method of preset initial value.Finally,by analyzing the data errors generated when processing the standard SIR infectious disease model,an innovative time-varying weight reserve pool calculation model was proposed.The results show that the time-varying weight model can effectively improve prediction accuracy compared to single level RC,and can reduce errors by 45.6% on real data test sets.Thirdly,in response to the shortcomings of the random matrix inside traditional RC,which leads to variable output results and long preprocessing data,the Next Generation Reservoir Computing(NG-RC)is used for further research on epidemic prediction.Firstly,an optimization structure is proposed based on the inherent characteristics of the NG-RC algorithm,which solves the problem of computational difficulty in processing high-dimensional data and improves computational efficiency.Compared to the previous synchronous prediction,this section also proposes a time difference prediction method,which meets the demand for predicting the future development trend of the epidemic on the timeline.At the same time,this chapter also studied the effectiveness of cross state prediction between different data,and the specific results are elaborated in Chapter 5.To sum up,this paper starts from the nonlinear dynamic model of infectious diseases,and combines the calculation of reserve pool with the development of COVID-19 epidemic,achieving the effect of forecasting the development trend of COVID-19 epidemic.