A New Inference Algorithm of Dynamic Uncertain Causality Graph Based on Conditional Sampling Method for Complex Cases

Abstract

Dynamic Uncertain Causality Graph (DUCG) is a recently developed model for fault diagnoses of industrial systems and general clinical diagnoses. In some cases, however, when state-unknown intermediate variables are many, the variable state combination explosion may appear and result in the inefficiency or even disability in DUCG inference. Monte Carlo sampling is a typical algorithm to solve this type of problem. However, since the calculation values are very small, a huge number of samplings are needed. This paper proposes an algorithm based on conditional stochastic simulation, which obtains the final calculation result from the expectation of the conditional probability in sampling cycles instead of counting the sampling frequency. Compared with the early presented recursive algorithm, the proposed algorithm requires much less computation time in the case when state-unknown intermediate variables are many. An example for diagnosing Viral Hepatitis B shows that the new algorithm performs 3 times faster than the recursive algorithm and the error ratio is within 2.7%.