Abstract: This paper presents a prediction model of selfconfiguring Artificial Neural Network(ANN) based on backpropagation learning (B-P) algorithm which is well known in the artificial neural network literature. So far, most studies on learning characteristics of B-P models have been concentrating on adjusting the connection weights of neural cells in which the rules of learning and connection mode remain unchanged. Considering many factors, such as structure and selflearning configuring intensively affect the properties of B-P models， this paper established a new selfconfiguring algorithm which can adjust the structure automatically by means of increasing or decreasing the number of hidden nodes. The model can also modify the learning rate of the network automatically in accordance with the value of output error which controls the learning rate. When the output error becomes large, the learning rate is also enlarged, then the enlarged learning rate leads to the small error output， so the convergence speed is improved and the system oscillation is avoided. Furthermore the time series were used as the model couples of input and output of artificial neural network, and the program about constructing prediction model was introduced． In order to simplify the algorithm and speed learning rate, the input data and sample data were processed by a special method. The results showed that the prediction values based on the model are in good agreement with the original values by referring the maximal relative error.