RESEARCH AND APPLICATION OF PSO-BASED COMBINED FORECASTING GRAY VERHULST-BP NEURAL NETWORK MODEL

The gray Verhulst model has the extremely widespread application in the study of minority, poor information and uncertainty question when the data show saturated state or s-shaped sequences. However the gray Verhulst model built by weakening the randomness of data sequence, lacking of self-organizing and self-learning. Some scholars study on this issue, and put forward a kind of gray Verhulst-BPNN combination forecast model. In this model, Partial-data set is used to establish Verhulst model group and BP neural network is utilized to build up the nonlinear mapping between partial-data Verhulst model group and original data in order to overcome the defects of the neural networking training with small sample of time series data. However, gray Verhulst-BPNN combination forecast model still has the problems of the local minimum and slow convergence caused by adjusting the network connection weights with error back propagation. Considering the PSO algorithm has the advantages of high accuracy and fast convergence, this paper put forward a kind of PSO-based combined forecasting gray Verhulst-BPNN model. Experiments show that the combined model has higher prediction precision and good stability.