The Reaserch Of Recursive Forecasting Method For Elevator Traffic Flow Based On SARIMA

Recursive forecasting method for elevator traffic flow based on Seasonal ARIMA (SARIMA) is studied in this paper.Proper forecasting for elevator traffic flow in elevator group control algorithm could provide useful prospective guidance for dispatching unit of elevator groups. In this thesis, based on detailed analysis of elevator traffic flow, according to the fact that elevator traffic flow series is just a univariate time series, we proposed a recursive forecasting method based on SARIMA.Based on reading literatures home and abroad widely, and in view of the questions in which elevator traffic flow forecasting method exists, a method that unifies off-line time series analysis and on-line Kalman recursive forecasting was proposed in this paper. In order to reduce the influence of abnormity value included in the training data for the forecasting, we introduce the outlier detection to off-line time series analysis. This method includes two steps: off-line analysis and on-line forecasting. In the first step, we analyze elevator traffic flow data using seasonal time series and get a initial time series model, then, start the outlier detection module and get corrected series according to outlier series, we can get the finial seasonal time series model of elevator traffic flow at last; In the second step, we transform the finial seasonal time series model to state space model firstly, and then, adjust the parameters of models using Kalman filter, and realize the on-line forecasting. Because we use SARIMA model as the foundation, the initial value of state vector in the Kalman filter is close to the correct value, so that reduce the numbers of iteration and the runtime. At the same time, the result of forecasting is more exact because of the outlier detection in the off-line analysis. All of these is the innovator of this paper.A case of simulation is completed by the environment Matlab 7.0 and SAS 9.0. The result and contrasts of various simulation experiments show that the forecasting method proposed in this paper can receive preferable result on both runtime and precision.