Fusion based estimation of the a-priori probability distribution of unknown non-stationary processes

Non-stationary processes can be hard to handle, particular if one would like to know their characterizing time dependent probability functions. In this paper the a-priori probability distributions of unknown non-stationary processes are estimated with different combinations of weakly coupled sensors. For quantification of the unknown a-priori probabilities Bayesian Networks (BN) are adopted for data fusion and Dirichlet functions are applied on non-stationary, time-dependent maximum likelihood (ML) parameter learning. In several experiments the adaption of the non-stationary a-priori probability density functions is shown and the accuracy of data fusion regarding the underlying process variables with different characteristics are determined quantitatively. It is shown that the proposed algorithm can improve data fusion in case conditions for specific process and sensor characteristics are met.