Title:
【発明の名称】神経回路網原子力プラント診断装置
Document Type and Number:
Japanese Patent JP2699607
Kind Code:
B2
Abstract:
PURPOSE:To diagnose a nuclear power plant after learning without any knowledge base by utilizing a neural network which can be calculated from the coupling between the output of a neural element and outputs of other neural elements including itself. CONSTITUTION:The neural network consists of neural elements in N input layers 40a, L intermediate layers Z(40b), and M output layers 40c and the neural elements 40 are simulated by neural cells of a living body. The neural elements 40 are connected by coupling elements 41 and the respective outputs of the neural elements are multiplied by coupling weight and outputted to neural elements in next layers to simulate synapsis. Here, N is the sum of the number Na of alarm states obtained from the nuclear plant, the number Nb of pressure values, and the number Nc of temperature values, and M is the number of accident causes. The coupling weight is calculated by learning algorithm from learning data on measurement data, the alarm, pressure, temperature, etc., obtained by simulation, etc., the accident causes, etc. After the learning, the accident causes are estimated actually for the alarm and others to diagnose the abnormality of the nuclear plant without any knowledge base.
Inventors:
Yoshio Izumi
Application Number:
JP8952090A
Publication Date:
January 19, 1998
Filing Date:
April 03, 1990
Export Citation:
Assignee:
Mitsubishi Electric Corporation
International Classes:
G21C17/00; G02F7/00; G06E3/00; G06F15/18; G06N3/00; G06N99/00; (IPC1-7): G21C17/00
Domestic Patent References:
| JP2206721A | ||||
| JP392795A | ||||
| JP3154896A |
Attorney, Agent or Firm:
Kaneo Miyata (3 outside)