| 摘要 |
PROBLEM TO BE SOLVED: To correctly predict a large fluctuation by making a learning coefficient to a learning data with a large fluctuation large and proceeding the neural network learning. SOLUTION: The learning device is constituted of a time sequence data classifying part 10 classifies inputted time sequence data A1 -An (n>=1), a learning control part 20, a feedforward-type neural network 30, an inspecting part 40 and a predicting part 50. In the neural network 30, learning is executed by emphasizing a learning set dealing with a large fluctuation by utilizing that the change quantity of weight per one time is made to be large through the use of a parameter being the learning coefficient. That is, at the time of learning in the neural network 30. A large learning coefficient (a) is used in a learning data part set L with a large fluctuation width and the learning coefficient (b) being smaller than the learning coefficient (a) is used in the part set S with a small fluctuation width so as to execute learning. Therefore, correct prediction is enabled at the time of predicting the fluctuation by the predicting part 50. |
