Complicated characters of noise in dynamical measure data are discussed in this paper.The analysis results show that the complicated characters of noise comes from two essential properties of model error.One is complex error

which cannot be described by general statistics model

and another is uncertainty of systematic error.It is built that the mathematic model of including complex error described by optimal knot spline function and time varying AR noise.The model can separate accurately not only the error with secondary low frequency but also the time varying stochastic error from complex error data.Furthermore

data fusion model with systematic error state is presented and precision discount factor method based on the uncertainty of systematic error state is also realized.At last

the precision integrate evaluation process is given.Computing results show the method is adaptive to actual engineering situation well.