Multi-Innovation Stochastic Gradient Algorithms For Multivariable Output Error Type Models

Compared with the single-variable systems, the multivariable systems are charactered by strong coupling and heavy interference, so the multivariable systems described the characterstics of the industry process more accurately and completely, and more similar to the control-systems than the single-variable systems; however, the parameter identification of the multivariable sys-tems is much more difficult than that of the single-variable systems because of the strong coupling of the multivariable systems. therefore, the researches of identification for multivari-able systems are of great importance. The multi-innovation stochastic gradient algorithms for multivariable output error category models are proposed based on a large number of reference, and the simulations to test the algorithms is also conducted. The achivements of the thesis are as follows:1. An auxiliary based stochastic gradient algorithm is obtained for the multivariable output er-ror models. In order to achieve a fast convergence rate, an auxiliary based multi-innovation stochastic gradient algorithm is proposed according to the multi-innovation identification theory. The simulation results show that the performance of the auxiliary based multi-innovation stochastic gradient algorithm is much better than other algorithms. And the research also find that performances of the proposed algorithm are determined by the length of innovation.2. For the multivariable output error moving average models, the colored noise is expand into the the information vector, and the value of the colored noise is got by estimating. and then the auxiliary based extended multi-innovation stochastic gradient algorithm is deduced. The simulation not only demonstrates the effectiveness of the algorithm, but also find that the performance of the algorithm is better for the lower-order models.3. For the the multivariable output error autoregressive average models, the output of the noise models is estimated, and then the auxiliary based general multi-innovation stochastic gradient algorithm is deduced. The simulation shows the effectiveness of the algorithm, and the research finds that:the performance of the algorithm is great even if the noise-to-signal ratio is great.4. For the multivariable Box-Jenkins models, by combining the auxiliary method and the general extended method, the auxiliary based general extended multi-innovation stochastic gradient algorithm is deduced. The simulation shows the effectiveness of the algorithm, and the research finds that:under certain conditions, the proposed algorithm works well.