This work studies the problem of dynamic data reconciliation through a nonlinear dynamic data reconciliation (NLDDR) code based on the dynamic optimization problem with nonlinear constraints associated to a certain calculation horizon. The algorithm is tested on a continuous stirred tank reactor (CSTR) polymerization reactor. A simultaneous strategy of solution and optimization is used to solve the optimization problem. For such a purpose, the optimization problem is turned into a nonlinear programming problem (NLP) through transformation of the ordinary differential equations of the model into a system of algebraic residual equations, which are introduced as a constraint on NLP. The transformation is made using the orthogonal collocation method on finite elements. Successive quadratic programming (SQP) is the technique used to solve the NLP problem, allowing insertion of equalities and inequalities algebraic constraints calculated on-line. To reach a good performance, same methods demand a higher number of samples, which increases the optimization problem dimension and turns the on-line numeric solution to nonviable. An appropriate technique manner to solve or attenuate such a problem is the implementation of calculation horizon along the operation time. Results emphasize the effects of such horizon in the methodology employed.