Research On Combination Forecasting Of Hospital Outpatient Quantity Based On EMD

The number of outpatients is an important indicator in the operation and management of hospitals,and it is of great significance to hospital management decisions and overall planning agreements.The influencing factors of outpatient volume are complex,and the direction and degree of action are difficult to quantify.It is difficult to predict outpatient volume through influencing factors.Therefore,this article analyzes outpatient volume data from the perspective of time series,builds a time series model to obtain hidden information of historical outpatient volume data,and accurately predicts the future outpatient volume to meet the needs of hospital health management.At present,in the prediction research in multiple fields,the continuous updating of models and prediction methods provides flexible options for complex and diverse time series prediction.The relationship between the prediction performance of the model and the data is inseparable.Models with different properties can reflect the data information from different aspects.Therefore,this thesis first analyzes that the daily outpatient volume of a third-class A-level hospital is periodic and stable,and the monthly outpatient volume has a period non-stationary,and trending.According to the characteristics of the data,the autoregressive moving average(ARMA)model and the three-dimensional exponential smoothing model(Holt-Winters)were used for exploratory experiments and related discussions.Combined with previous research and the development of forecasting trends,a combined framework based on Empirical Mode Decomposition(EMD)was selected for forecasting research.An integrated decomposition and forecasting strategy was adopted to extract multi-scale features of complex sequences and reduce the redundancy of sequence information.At the same time,in order to obtain satisfactory results,the optimization study was first carried out under the original framework,and then a prediction model based on EMD fusion multi-long short-term memory network and neighborhood algorithm was proposed.The method uses EMD to decompose the outpatient quantity data sequence into several sub-sequences,select the fluctuation component and the overall component through the sample entropy of the sub-sequences,and establish a prediction model based on the neighborhood and multivariate LSTM neural network according to the characteristics of the components.The main research is divided into two parts:(1)Research on the problems existing in the EMD combined forecasting framework.In previous studies,the prediction errors caused by the defects of the EMD decomposition sequence were ignored.This article mainly discusses the end effect.Secondly,in the development of EMD combined prediction framework,the lack of full use of feature time scale information,only using single factor prediction,this article will discuss the construction and application of multidimensional feature time scale information.Based on the above two researches,this research uses feature matrix input and data extension methods to increase the accuracy of predictions under this framework.(2)Research on the specific details of the sequence 'divide and conquer' prediction strategy and model combination.Because the sequence complexity of the decomposed sequence is inconsistent,the sample entropy is introduced to classify the sequence,and the fluctuation range is large.The complex sequence uses long-short-term memory neural network.For the sequence with relatively stable changes,this article quotes the relevant knowledge of the recommended algorithm.Neighbor algorithm based on time information is used for estimation and prediction.Two sets of data of daily outpatient volume and monthly outpatient volume were selected as experimental data,and the existing representative prediction methods were compared.