Reliable Identification Algorithm Based On Least Absolute Deviation

In this paper,the reliable identification algorithm based on least absolute deviation is proposed innovatively.Reliable identification algorithm is defined as high-precision and strong-robustness identification algorithm.And research on reliable identification algorithm is also one of the important fields of control.For the case that the intelligent optimization algorithm does not require the objective function to be continuous,the gravitational search algorithm(GSA)based on the least absolute deviation is proposed,and some strategies are proposed to improve the parameter estimation accuracy and the identification speed.The GSA based on the least absolute deviation has a strong global search ability and local search ability;and it can still identify parameters accurately when the number of samples is small.When the model order of the multivariable closed-loop system does not satisfy the traditional identifiable conditions of the closed-loop system,there will be a linear correlation between the observed data vectors,and then the model parameters cannot be estimated.The multivariable closed-loop system identification algorithm based on approximate partial least absolute deviation proposed uses principal component analysis(PCA)to realize decorrelation in the above case,and obtains the unique solution of the model parameters.The algorithm converts the unidentifiablility problem into an identifiablity problem and suppresses the impulse noise effectively.The identifiability of the multivariable closed-loop system is proved theoretically.The final chapter mainly focuses on the multivariable Hammerstein model where linear correlation exists between the input signals and the nonlinear correlation between input signals.The purpose is to study reliable identification algorithms of nonlinear systems with superior performance.A multivariable Hammerstein model identification algorithm based on approximate partial least absolute deviation(MPALAD)and radialbasis function(RBF)identification algorithm based on approximate partial least absolute deviation(RBF-PALAD)are given.Both algorithms use approximate partial least absolute deviation as the main method to realize decorrelation.The RBF-PALAD algorithm expands the input matrix for the non-linear correlation model of the input signal.The extended term is the hidden-node output of the RBF network to deal with the nonlinearity of the system.Simulation experiments when white noise and impulse noise exist verify the reliability of the proposed algorithm.Compared with the algorithm based on least squares criterion,the proposed algorithms have better accuracy,stronger robustness and higher reliability.