The invention provides a matrix-decomposition-based method for realizing stateful data in Bayesian network. As the obtained row vector and column vector in the stateful mode based on matrix decomposition are corresponding to the state likelihood ratio of the prior data and evidence and the standard state in the Bayesian network, the integrity and accuracy of the inference data can be preferably guaranteed. The matrix-decomposition-based method includes the steps: during the process for constructing the Bayesian network, performing matrix decomposition on the prior data, thus obtaining the row vector of the state likelihood ratio of the prior data and the row vector of the standard state in the Bayesian network, analyzing and counting to obtain a condition probability table, and constructing the Bayesian network by means of integration with the causal relationship among the characteristic quantities; during the inference process, performing matrix decomposition on the evidence data to obtain the evidence data with the standard state, and then accessing the Bayesian network to perform inference fusion; and at last, combining the inference result in a matrix and restoring the data to obtain the final result.