Research On Predictive Model Based On Error Back Propagation Neural Network

Since the industrial revolution, industrialization drive people's accelerated pace of life, but also brought the rapid development of the market economy. In this background of rapid development, the survival of enterprises is also high-speed change or perish. As we all know, Android phones can rise within a short span of three years, ranked number one market share of more than a decade of Symbian, the world's first, but also for the majority of enterprises to provide a reference, so that people deep in thought. Enterprise development which way to go, how to succeed in the economic struggle, the sales forecast will enter people's attention, thereby predicting the subject becomes hot. Entrepreneurs are eager to provide basis for decision making by predicting the development of their own business, understanding of user needs, so that it can compete to win the economic war.In this paper, under the economic situation is so fierce competition, discusses the basic theoretical predictions disciplines, introduced a variety of forecasting methods and techniques of market forecasts, as well as information and data which they are adapted, for the majority of entrepreneurs, enabling them according to their own master data, to make the next step into, sale, deposit strategic plan. Due to the traditional single predicted to have low accuracy, will lose some data and other shortcomings, we use a combination of methods to predict thinking elaborate, creates a time series model of the seasonal cumulative ARIMA model(SARIMA) and Error back-propagation neural network forecasting method combinations. This is so chosen because the sales data has the characteristics of time series, and the factors that affect sales but also diverse non-linear model to use SARIMA get fit residuals, and then use the back-propagation neural model powerful nonlinear processing ability to predict and then to improve the prediction accuracy. Since conventional back-propagation neural network convergence speed is slower, this article by dynamically adjusting the connection weights of neurons in different ways between nodes to improve the convergence speed of neural networks to obtain better results.Finally, Lenovo laptop from 2005 to 2014 sales data for the study, the use of combination forecasting method proposed above for future sales forecast, as sales forecast for other enterprise systems to provide a reference, and to help enterprise managers to develop sales plans reasonable.