Predicting The Incidence Of Bacillary Dysentery In Shanghai Based On The BiLSTM-GRNN Combined Model

In December 2019,the outbreak of pneumonia caused by the new coronavirus in many places,such as Wuhan,has greatly affected China's economic development,social order and other aspects.In February 2020,outbreaks of infectious diseases also broke out in Brazil,Nigeria and other places,triggering world citizens'"outbreak panic".The above public health incidents brought the concept of prevention and control of infectious diseases back to Public view.Many experts and scholars from all over the world have done a lot of research work on infectious disease prediction,and have also achieved many remarkable achievements,such as wavelet model,grey dynamic model^[36],ARIMA model^[23-26],etc.,which have been successfully used in many aspects.But at the same time,these predictive models also have various disadvantages,such as high sample data requirements,high computational complexity,and poor stability.As the incidence of infectious diseases is affected by many factors with obvious randomness and volatility,it is difficult to get an ideal high-precision prediction results using general linear regression.With the continuous deepening of machine learning and deep learning research,many non-linear problems with high randomness and unpredictability can be better solved,and the final model fitting accuracy is more ideal than that of ordinary general linear models.This paper focuses on the research work of intelligent prediction models based on the combination of BiLSTMand GRNN.The data of bacillary dysentery?BD?in Shanghai from 2004 to 2019 was used as the main research object.By comparing the advantages and disadvantages of this model with other contrastmodels,its applicability and effectiveness can be discussed.The work focuses on the following two topics:Firstly,construct a new intelligent model.The main research work is divided into five aspects:?1?The four kinds of research data are normalized,and the dimensions of the research data are unified before modeling;?2?The two algorithms selected in this paper are analyzed on the deep learning architecture for effective combination,the feasibility of constructing a double-layer network structure,?3?Determining the best deep learning parameters,?4?Using the pre-processed data for model fitting,?5?Comparing the prediction comparison chart of different training times with the Loss chart to verifythe prediction of the new intelligent model in the incidence of BD Effectiveness.Secondly,fit the contrast model and verify the high accuracy advantages of BiLSTM-GRNN horizontally.CS-LSTM and LSTM-GRNN were selected as contrast models.After modeling calculations,the MSE,MAE,and RMSE of the BiLSTM-GRNN prediction model were?4.205e-05,0.0062,0.0064?after the same number of iterations,which were lower than CS-LSTM.?0.0014,0.0356,0.0356?and?0.0002,0.0152,0.0151?of the joint model with LSTM-GRNN,indicating that comparing with the neural network optimized by an optimized algorithm and the dual neural network with unidirectional error correction,the closed-loop dual neural network with error correction's prediction accuracy is higher,the fitting result is better.Finally,in order to verify the robustness of the new model,Friedman's significance test was performed on the three groups of models.After calculation,it was found that there was a significant difference between the residuals of the BiLSTM-GRNN combined model and the contrast model,indicating the robustness of the new intelligent model is higher.Through the modeling and research of the BD incidence sequence in Shanghai,a more suitable predictive model for the incidence of infectious diseases was explored,which made up for the lack of statistical prediction research on infectious diseases,and provided the relevant digital and analogue support for the prevention and control of intestinal infectious diseases in China.The research results have good application value.