| 摘要 |
PURPOSE:To efficiently and speedily adapt to change of a situation by constituting a neural net using predictive waiting time as output at a first stage, appropriating an input layer and a middle layer at a second stage, learning with an educator signal by way of only calling an output layer and reconstituting it for assignment and performing reinforced learning by way of limiting it only to the output layer at a third stage. CONSTITUTION:At a first stage, a neural net using a system situation data of a group as an input signal and information closely related to call assignment and capable of acquiring a highly precise educator signal as an output signal is constituted and learning is performed by using the educator signal. At a second stage, by leaving an input layer and a middle layer of the neural net which complete learning at the first stage as they are and reconstituting the output layer for the call assignment, the output layer is learned with the assignment result of a conventional call assignment method as the educator signal. Thereafter, at a third stage, the neural net which finishes with the second stage is applied to a working site for the call assignment, and only the output layer is appropriated to the working site at the reinforced learning during operation of an elevator. |
