| 摘要 |
PURPOSE:To discriminate quickly the cause by which an abnormality has been generated, by executing automatically and simultaneously a learning and a diagnosis by a pattern vector having the highest sensitivity with respect to an unknown abnormality of a plant. CONSTITUTION:When a diagnosis start command of a nuclear power plant is given, its various statistical quantities are brought to a fast Fourier transform and analysis as one block at a time interval DELTAt, and also a statistical quantity Yi is calculated as an average of its N block. Subsequently, a sample pattern vector Xi is constituted by setting an initial condition, and a state Ki is determined by calculating a reference vector of the same constitution, which has been learned by a trial of said vector and a state K. Also, an RMS value of a frequency component of the outside of an area is calculated, and a distance between this logarithmic converting value and a value in a normal state is calculated, and a proportion to a threshold level is decided. A plant state Ko is determined by repeating said procedures, and at the time of an abnormal state, a new pattern vector is constituted, it is made a new reference data set, an alarm is raised, also a display is outputted and the diagnosis is ended. |
