連続測定データに対する確率的定性推論の高速化

抄録

A stochastic qualitative reasoning method has been developed for fault diagnosis of building air condition-ing systems. In this method, the fault part can be identified by ‘agreement rate’, which is a parameter that shows how reasoning behavior reflects observations. However, when agreement rate is calculated at a certain point of time, reasoning must be done with several unit times from that point. Accordingly, it is inefficient to do reasoning several times at each point.
In this paper, a continuous qualitative reasoning method is proposed. In this method, reasoning is done from the beginning of observations to the end. In addition, a technique for estimating the minimum exis-tence probability, which is used to remove the states with small occurrence, is introduced. As results, the performance has increased about 40 times in comparison with previous method.