Extreme value analysis has been widely used to assess and predict environmental catastrophes caused by climate change over the years. In addition to the environmental area, other common areas of application of these analyzes are finance, actuarial, among others. In this way, the present work consists in the estimation of parameters and levels of expected returns, considering the distribution of extreme values for the $r$-highest order statistics. These estimates will be evaluated in series that have points of change in the regime, that is, a model will be proposed to detect points of change in a series, applied to the distribution of the r-largest order statistics (GEV$_{r}$). We will address the case where the series has $k$ change points, in which the series has $k+1$ different schemes, and each scheme will be modeled by the GEV$_{r}$ distribution. The inference used in the model is based on a Bayesian approach, where both the GEV$_{r}$ parameters for each scheme, and the change points are considered as unknown parameters to be estimated. In addition, an evaluation of the criterion of choosing the optimal $r$ for the distribution of the data. The estimation is performed by the Monte Carlo Method via Markov Chains (MCMC) using the Metropolis-Hastings algorithm technique. Initially only simulations were performed, regarding the value of fixed and iterative $r$, in which remarkable results were obtained. In addition, a brief analysis of levels of returns for different values of $r$, and a brief descriptive analysis of the real data that will be used in the applications of the proposed model, was carried out. Finally, the application of the proposed model for the data of quotas of the river Parnaíba, a Brazilian river that bathes the states of Maranhão and Piauí.