| 摘要 |
<p>PURPOSE:To estimate the application limit of each component such as a sensor in a real time and to lessen uselessness in estimation by learning a plurality of patterns of changes with time in characteristics of a component of the same kind which are selected from information on the changes with time stored in a storage means. CONSTITUTION:Information on changes with time in characteristics of a component of each kind in various states of operation of a plant is collected, and a plurality of patterns of changes with time in the characteristics to be learned by a neural circuit model are selected from this information on changes with time. With these patterns of changes in the characteristics in a plurality used as input signals and with the assortments of these patterns of changes in the characteristics used as supervisor signals, learning by the neural circuit model is executed. Patterns based on information on changes with time till the present time of a component of which changes in the characteristics are to be estimated are inputted to the neural circuit model having made this learning, the degree of similarity of the inputted patterns to the discrete patterns of changes with time in the characteristics having been learned is determined from outputs of the neural circuit model for the inputs, and based on the degree of similarity thus determined, future patterns of changes with time in the characteristics of the component are estimated.</p> |
