Nuclear power plants transient diagnostics using LVQ or some networks don't know that they don't know

摘要

A nuclear power plant's (NPP) status is monitored by a human operator. Any classifier system used to enhance the operator's capability to diagnose the NPP status should classify a novel transient as "don't know" if it is not contained within its accumulated knowledge. In particular, a neural network classifier needs some kind of proximity measure between the new data and its training set. Multilayered perceptron (MLP) networks do not have that measure, while Kohonen self-organizing maps (SOM) and learning vector quantization (LVQ) networks do. This measure may also serve as an explanation to the network's decision the way case-based reasoning expert systems do. Applying an "evidence accumulation" technique by using a transient's classification history can enhance the network's accuracy as well as its consistency.