Partial least squares (PLS) regression has been effectively used as one of the data-driven empirical modeling to deal with a large number of variables in the complicated covariance structure. But Partial least squares (PLS) regression has limitations in the online model update. Therefore, recursive PLS has been used for the online adaptation of the model. The RPLS algorithm is implemented as soon as some new samples are available. It may be desirable not to update the model until significant amount of data are collected and the process has gone through significant changes. In this paper, we used the blockwise recursive partial least squares (RPLS) algorithms with a moving window for the adaptation of the model according to the process shifts. The block-wise RPLS algorithm has been used to reduce computational load in PLS regression and its cross-validation. In this work we defined the index to represent correlation between blocks and proposed how to determine the forgetting factors through it. This proposed method is to improve the prediction power of the adaptive modeling considering correlation between blocks without general process knowledge. Therefore, this proposed method included correlation information that different methods did not express. The method was tested to process heaters in the oil refining company. We constructed the prediction model of the effluent NO_x composition with forgetting factors and showed that the prediction power approximated more than 95% of the minimum error.