In order to solve the joint probability distribution of Multivalued Dynamic Uncertain Causality Graph with domain knowledge, it is difficult to construct the probability distribution expression of domain causal graphs. An inference algorithm based on graph decomposition is proposed from two kinds of cyclic causal graphs, which greatly simplifies the construction of global probability distribution expression, effectively reduces the difficulty of constructing probability distribution expression of domain causal graphs, and improves the inference efficiency of multi-valued dynamic uncertain causal graphs.