Adaptive Model Recovery Scheme for Multivariable System Using Error Correction Learning

In automatic control design, identifying the system parameters based on an effective identification algorithm is often necessary. Given that many realistic plants are a multivariable system, such identification is considered critical. Most of the available multivariable system identification techniques are designed based on prediction error-correction learning, which produces an unsatisfactory estimation accuracy and low convergence speed, especially in the case of strong external interference. In this article, an adaptive identification scheme is proposed to achieve model recovery for a multivariable system based on a novel error correction learning framework. First, three fictitious sub-models are established by means of the hierarchical principle, in which the high computational burden of the identification approach is avoided. Second, the aforementioned identification method is proposed to recover the parameter information of each sub-model. To improve the identification performance, the identification error information contained in the system data is derived and used to establish a criterion function. According to the identification error and initial parameter error data, a novel criterion function structure relying on the regularization and punishment mechanisms is proposed. Based on this criterion function, a new adaptive error correction learning parameter update law is then deduced. A numerical example and a real-world plant are examined to verify the advantage and practicality of the presented adaptive identification scheme.