Weight Self-Learning Based Combination Forecasting of Product Diffusion

This paper deals with a weight self-learning approach to combination forecasting for the product diffusion process. Today, technology develops so fast that development cycles frequently exceed the market life of new products. The forecasting of product diffusion should be adaptive to different stages of diffusion process. The different product diffusion models are combined to form a model base, which can be expanded and amended by the combination of different models for the demand of forecasting product diffusion behavior in the different stages. The product diffusion model base is denoted as weight matrix, on the basis of which a self-learning algorithm for the product diffusion behavior forecasting is advanced. The convergence of the algorithm is analyzed. With the algorithm, the weight matrix is amended by combined forecasting and self-learning method, and the progressive forecasting of the product diffusion behavior is realized. An example is given to testify the validity. The results show that the self-learning algorithm is of both long-term and short-term accuracy.