The Two Parameters Estimation Methods And Application Of Variable Coefficient Model For The Second Half Under The Constraint Condition

Half a variable coefficient model is the effective promotion of variable coefficientmodel, it is part of the organic integration of the advantages about the part of linear modeland the variable coefficient model, it has greater flexibility and applicability than the partof linear model and the variable coefficient model. The unconstrained problem of theconjugate gradient method is a kind of very effective methods in solving unconstrainedoptimization, this paper re-builds and applies it to half varying coefficient model, it notonly retains the advantages of fast convergence of the conjugate gradient method, but alsobecause of advantages of leaving makes the multiplier effect of superposition.This paper builds two conjugate gradient projection parameter methods of half avariable coefficient model under the constraint conditions, this paper also gives theARMA model parameters estimation methods and application.Firstly, this paper discusses half a variable coefficient model in error is ARMA (1,1)sequence the estimation problem of the function coefficient α (·), this paper also builds aP-CD conjugate gradient method projection in the linear equation under the restriction.This algorithm is applied to the half of the variable coefficient model parameter estimation.Then the algorithm for the decline of sex and convergence is proved, with the examplesshowed that the proposed algorithm is more efficient.Secondly, this paper builds up a kind of double parameters of the conjugate gradientmethod, and this method is applied in half a variable coefficient model parametersestimation. Simultaneously, the decline and convergence of the algorithm are proved, alsothe example of this algorithm was tested, the results show that the effect is good.Finally, this paper gives the application about the BFGS algorithm in ARMA modelparameter estimation. This method shows that the accuracy is better in the parameterestimation and prediction performance on the basis of the algorithm's global convergence.