This work demonstrates the development of a practical predictive control approach to regulate the outputs of a network of heat exchangers. The full nonlinear continuous-time dynamic of the network is introduced to simulate the real process. Whereas a discrete linear and simpler version of this dynamic is employed to construct the predictive controller as well as an extended Kalman estimator. The presented algorithm is a time-varying model predictive control (MPC) where the simplified model is successively linearized on-line around the current available measurements. For the closed-stability of the introduced MPC scheme, a modified method is suggested to design terminal stabilizing weights where reference tracking problem, external disturbance dynamics, as well as the proper dynamics of a heat exchanger with bypass are all considered. The on-line updated extended Kalman estimator is utilized to estimate both the process states and unknown disturbances. Thanks to the successive linearization technique, the introduced predictive controller benefits from efficient online computations and online adaptations in operation points. Nonetheless, an accurate model of the heat exchanger is essential. Compared with a typical linear MPC, it is concluded that the presented predictive algorithm produces satisfactory results for disturbance rejection and switched operation points circumstances.