An Improved Variable Learning Rate Backpropagation Algorithm And Its Application In Sales Forecast

Artificial neural network is a nonlinear system which simulates biological neural systems. It has powerful parallel processing, distributed storage, adaptive learning, a high degree of fault-tolerant capability. Dozens of neural network models have been proposed for decades and has been widely applied in many fields. Owing to simple structure and technical mature, BP network became the most widely used network. However, BP algorithm uses the steepest descent algorithm, thus there are many shortcomings, such as local minimum problem, slow convergence rate and Initial value affected the convergence.In this paper, We recviewed neural network structure, classification and learning algorithm, and then discusses the BP neural network model, learning methods, and introduced the status of BP improved algorithm.Learning rate plays an important role in BP algorithm. This paper studied learning rate and proposed an improved BP algorithm based on VLBP. The proposed algorithm connects the learning rate with the error ratio. The proposed algorithm introduce a function used to dynamic adjust the learning factor. The independent variable of this function is error ratio between two calculate. Through the XOR problem, approximation to nonliner function problems,breast cancer pattern classification problem and function approximation problems with noise, simulation results demonstrate that the improved VLBP algorithm could effectively improve the rate of the algorithm convergenceFinally, An artificial network model for sales forecasting of mobile phone is proposed in the paper. By applying the improved VLBP algorithm into the sales prediction,this paper show superiority of improved VLBP algorithm after compared with VLBP algorithm and other methods in popularity. Experimental results are presented to show the effectiveness of the improved VLBP algorithm and responsibility to the fact.