A data-driven real time optimization (RTO) scheme is proposed for complex processes under uncertainties. The objective is to optimize the economic performance of a plant when its first-principle model is difficult to obtain. To this end, a piecewise affine (PWA) model is built by fitting the operating data to predict the production of a process. The trust region of the PWA model is also characterized to ensure reliable predictions. Then, the RTO is formulated as a mixed-integer nonlinear program (MINLP) based on the PWA model. To alleviate the computational demands, we also develop a dynamic programming (DP) scheme that solves the RTO to a near-optimal solution. Two case studies show that the PWA model can accurately forecast the real productions when the process is operated within the trust region, and the proposed DP scheme takes less time to find a better solution than the state-of-the-art solver for the large-scale MINLP. (C) 2019 Elsevier Ltd. All rights reserved.