The Research Of Multi-mode Forecasting Method For Elevator Traffic Flow

Multi-mode forecasting method for elevator traffic flow is studied in this paper.As known, proper analysis on elevator traffic flow could provide useful andprospective guidance for dispatching unit of elevator groups, which draws peoples'attention to the problem of elevator traffic. In this thesis, based on diversity ofelevator traffic, a hybrid multi-mode forecasting approach is proposed. Firstly, theelevator traffic flow is recognized and classified into patterns through ArtificialImmune Clustering Algorithm (AICA), and then Gaussian Mixture Model (GMM)models the multi-pattern elevator traffic flow to predict future value on-line.During the process of research, after deeply discussing the characteristics andapplications of various prediction methods, AICA is applied to cluster elevator trafficinto eight traffic patterns particularly and calculate the proportion of each trafficpattern, which improves the accuracy of the prediction model and also avoids thelimits of four traditional traffic patterns: off-peak, inter-floor, up-peak and down-peakpatterns. Therefore, the qualitative analysis on elevator traffic has been accomplished.Afterwards, GMM, whose parameters are estimated on-line with ExpectationMaximum (EM) algorithm, is applied to approximate the transition probability densitydistribution of elevator traffic flow. The number of Gaussian components is equal tothe number of traffic patterns, and the initial mixing coefficient of each component isequivalent to proportion of traffic pattern after being clustered. With a quantitativemodel, the future value of traffic flow is predicted. Combining AICA with GMMincreases the precision of forecasting, and also offers a perfect beginning for EMalgorithm to reduce the iteration steps and forecasting period. So, the most innovationin this thesis is integration of qualitative analysis and quantitative analysis.A case of simulation is completed by the software environment of Matlab6.5.The results and contrasts of various simulation experiments show that the multi-modemethod can predict the future of elevator traffic flow exactly.