Research On Time-varying System Identification Algorithm

The time variant parameters widely exist in many fields such as process control, avigation and spaceflight, fault diagnosing, robot control. As the extensiveness and complexity of time-varying systems, the limitations of accessment to information from time-varying systems, the identification and control focus on the time variant systems already becomes a hotspot of the systematic science and control science. Traditional algorithms focus on invariant variables are proposed based on the hypothesis of time invariant system or stationary stochasitc processes, but the algorithms exist certain limitations because of the actual signal and process always present various kinds of fluid processes and time variant.This dissertation studies identification algorithms of time-varying system which based on abundant correlative literatures about linear time-varying system parameters identification, with certain theoretical and potential values in applications.The main work is summarized as follows:(1)The constant forgetting factor is used in linear time-varying systems identification frequently, but the constant forgetting factor has the disadvantage in tracking time variable parameters. A kind of bounded auto-regulation forgetting factor is proposed to replace the constant forgetting factor for identifying ruleless parameters change. The new time-varying forgetting factor is regulated by posteriori error, meanwhile limited the upper and lower bound by introduing a parameterσand the innovation variation in order to avoid the interruption's influence to the forgetting factor. The simulation results demonstrate the new parameters estimation algorithm possesses the strong ability for time-varying parameters and is able to overcome the constant forgetting factor's disadvantage in tracking time variable parameters.(2)A new algorithm which possesses modified criterion function is proposed, the algorithm is focused on time-varying OE model based on the generlized criterion function and the prior knowledge. Even if the input signal does not meet the conditions for persistent excitation, the algorithm is still be applicable in time-varying systems of the parameters identification through the introduction of adjusted parameterα, ensuring the effectiveness of the algorithm. The algorithm values small amount of computing and good convergence, providing the flexibility for obtaining time-varying parameter estimates and overcoming the ill condition, it is especially suitable for the parameters identification of few observed issues. The simulation results show the algorithm based on the modified criterion function performs better parameter estimations in linear time-varying system identification.(3)To solve the least squares'shortcoming in parameter estimations burst off which exist when covariance matrix's reduction, a new identification algorithm introduces both the damped factor and forgetting factor, effectively overcomes the parameters burst off and be able to track time-varying parameters. By introducing a damped factor, the parameters'incremental changes are added to value the relative importance in criterion function.