| 摘要 |
<p><P>PROBLEM TO BE SOLVED: To provide an apparatus, a program and method for predicting traffic which predict the upper limit value of traffic variation. <P>SOLUTION: The apparatus has: a result traffic data base in which a past result traffic and its date attribute information are assigned and accumulated; and a weight coefficient storage part which accumulates weight coefficients w<SB>i</SB>in an objective function. The apparatus has: a neural network processing part which performs regression processing by a neural network with minimizing of a weighted squared error as the objective function by using an error reverse propagation method and which outputs a learning traffic; and a weight coefficient deciding part which increase the weight coefficient w<SB>i</SB>according to a difference between a result traffic y<SB>i</SB>and a predicted traffic z<SB>i</SB>only when the result traffic y<SB>i</SB>is larger than the predicted traffic z<SB>i</SB>. The apparatus is controlled to repeat processing of the neural network processing part and the weight coefficient deciding part a prescribed number of times for each result traffic y<SB>i</SB>to calculate a weight coefficient becoming a prediction model. <P>COPYRIGHT: (C)2007,JPO&INPIT</p> |
