Research On Hospital Revenue And Expenditure Forecasting Method Based On Improved LSTM

Hospital revenue and expenditure is not only an important index that directly reflects the profit situation of the hospital,but also an important part of the analysis of the implementation of the hospital financial budget.The prediction of income and expenditure is of great significance to the operation status and future trend of non-profit hospitals,and is also the best way to promote the reform of public hospital management system and improve the operation mechanism of public hospitals.Among the current time series forecasting methods,ARIMA model has its own advantages in linear feature extraction and LSTM model in nonlinear feature capture,but they are both specialized in the extraction of a certain part of feature relations.Therefore,an income and expenditure forecasting method based on the integration of ARIMA model and LSTM model is proposed.The research work of this paper is as follows:Firstly,the characteristic components of the income and expenditure data are analyzed and the data composition is discussed.At the same time,ARIMA model and LSTM model are established to extract linear and nonlinear feature relations respectively.Due to the seasonal trend of the data,the combined model is further improved on the basis of the base model.This makes a further contribution to the accuracy improvement of the model.Secondly,in order to make full use of the linear and nonlinear characteristic relations in the data,this paper proposes a SARIMA-WLSTM combined model.In the process of budget execution analysis,the SARIMA model with seasonal processing is used to solve the linear relations in the time series data.The improved particle swarm optimization LSTM model was used to capture the nonlinear feature relation in the time series data.Based on the existing experience analysis,the weight ratio of the feature relation was carried out by using the calculation formula of error value,and the SARIMA-WLSTM combination model was established.Thirdly,in order to further describe the complex feature relationship in the income and expenditure time series data,improve the subjective problem of the weight ratio of the combination model and improve the prediction accuracy of the combination model,a feature fusion model R-SARIMA-LSTM based on the residual and neural network self-learning feature relationship is proposed.The residual error outside the linear feature relation extracted by SARIMA is used as the input of improved LSTM network to extract more accurate nonlinear features.Then,the linear feature relationship extracted by SARIMA was combined with the residual sequence prediction results,and the new LSTM network was used to self-learn the relationship between linear and nonlinear features,and the final prediction results were obtained.Finally,two different combination models are verified by examples.The results show that the optimized SARIMA-WLSTM model can reach 87.35% and 85.85%,while the RSARIMA-LSTM model can reach 89.49% and 87.91%.It can provide reference for actual forecasting work.