Automatic Construction of Linear Stochastic Dynamic Models For Stationary Industrial Processes with Random Disturbances Using Operating Records

摘要: We describe a new technique for automatic identification of stationary, linear systems with single output. This class models includes all linear, time-invariant, stochastic, difference equations driven by arbitrary inputs and having normal disturbances rational spectra.The parameters the model are estimated method maximum likelihood. A numerical algorithm solving likelihood is presented. The essentially modified Newton-Raphson algorithm, which takes advantage particular structure problem.Conditions consistency asymptotic efficiency estimates given increasing sample length. It shown that these properties exclusively determined information matrix. An estimate latter obtained without additional computations. matrix also yields an accuracy in each. case.The approach has been tested on artificially generated input/output data. immediately applicable to power spectrum analysis time series, advantages over ordinary non-parametric methods it always gives non-negative problems trend removal choice spectral windows turning up.The basic idea can be extended larger classes systems. Also easily done recursively, implies well suited real modelling. (Less)