A Sequential Projection Algorithm for Special-Services Demand
01 January 1982
A Sequential Projection Algorithm for SpecialServices Demand By A. IONESCU-GRAFF (Manuscript received December 31, 1980) Performance analysis of the four time-independent regression models presently used by Bell operating companies to forecast special-services circuit requirements, and the characteristics of actual special-services demand history observed from three operating companies, indicate a need for a new method to forecast these difficult time series. A new special-services demand sequential projection algorithm (SSD-SPA) is developed based on a linear Kalman filter model. It includes methods to detect previous deterministic events, to accept and process exogenous information affecting the demand, and to recognize and adapt to a "no-growth" situation. Compared to the present algorithm, SSD-SPA generates significantly better forecasts: approximately 30 percent improvement in forecast accuracy and stability, 25 percent reduction in rms error, and 22 percent reduction in circuit misplacements.