| Objective It is known to all, with China being joined the WTO, the exchanges of west-east being increasingly extensive, Western advanced technique and the management principle will inevitably impact domestic local market, and the structure and form of the domestic original hospitals will suffer the rigorous challenge. Then it must put on the pace of scientific hospital management modernization. And this also will require the statistician can make the most of the historical time series data of hospital to analysis and predict, and provide the quantitative forecasting results to hospital manager, which is regarded as the theories guide of the future work direction, avoiding the blindness of management work. The objective of this paper is to explore an appropriate and practical forecasting model for time series data of hospital, through forecasting the yearly data of outpatient numbers and the seasonal data of inpatient numbers.Methods Part One: In this study the data of outpatient numbers were collected from 1985 to 2002 in the some hospital in Tianjin City. The forecasting steps of the data were as follows: First, four different single forecasting models were given: cubic exponential smoothing model, cubic polynomial regression model, grey model and Box-Jenkins models. Then, it was set up a nonstationary weights mix forecasting model based on artificial neural network (ANN) with three layers. The inputs of ANN were the forecasting values of above four kinds of methods and output was the actual time series. At the same time, the result of the non-stationary weights mix forecasting model was compared with that of the stationary weights mix forecasting model, in contrast with the optimal weighted mix forecasting model.Part Two: In this study the data of inpatient numbers were collected from 1996 to 2001 by four seasons in the some hospital in Tianjin City. First, three different single forecasting methods were given: seasonal regression model, Winters' exponential smoothing method and Box-Jenkins models. Then ANN was applied to mix forecasting according to the results of above three models.Part Three: It was applied the health manpower data adopted from the article entitled: "Application of the optimal weighted mix forecasting model in the health manpower in Hebei province" in <> Vol.13, No.l, 2000. And the result of the mix forecasting model based on ANN wascompared with that of the optimal weighted mix forecasting model.Results The order according to the mean relative percent error of forecasting result from low to high was cubic polynomial regression model (0.0469), grey model (0.0655), Box-Jenkins models and cubic exponential smoothing model (0.0789). These single models were applied to build the optimal mix forecasting model. The combination results showed that forecasting precision was improved to some extent, relative error was 0.0543. But these methods could not pass the kendall consistencytest (x2=8.625, P>0.05), namely they did not have the consistency among thesesingle models, so we looked for another mix forecasting method. Finally it was constructed the mix forecasting method based on ANN to forecast the numbers of outpatient, relative error was 9.996 10-5.In addition, artificial neural network also seemed to provide adequate forecasts for the seasonal data of inpatient numbers, relative error was 9.988 10-5.Conclusion ANN has high self-study ability and can show the nonlinear relation among the data of sample with strong ability to approximate functions. And the weight coefficient of any single model in ANN is no longer a constant, but is a function with time. Although the influence that a single method work on the result of mix forecasting method is relevant to the forecasting results and weights of itself, it is a nonlinear relation. The mix forecasting method based on ANN overcame the limitation of stationary weights on optimal mix forecasting model, therefore it is a mix forecasting method of nonstationary weights in nature. It...
|
PDF Full Text Request