Sequential control has been applied to industrial batch reactors to track a predetermined setpoint trajectory, however, due to nonlinearities, unknown kinetics and the lack of on-line measurement of reaction rate, it is hard to obtain a uniform batch cycle time and a high product quality. Several temperature control methods are proposed for free radical batch polymerization reactors to improve the batch cycle time and the product quality. First, a fast fuzzy algorithm is proposed to implement a fuzzy controller in a distributed control system. The fuzzy controller using fast fuzzy algorithm which can handle the nonlinearites has a merit of easy implementation because of transforming the consequence and the defuzzification into an equivalent crisp function. And it is found that the performance of fast fuzzy algorithm is equal to that of Mamdani's algorithm, and superior to that of model based controllers. Also a systematic design procedure including fuzzy rule tuning is presented, which leads us to a fuzzy controller of which domain experts are not necessary. For rule tuning, the membership functions of the antecedence and the consequence, the input/output scale factors are tuned simultaneously by using a successive quadratic programming. Second, a predictive learning algorithm is proposed for the control of repetitive batch processes. By applying a prediction horizon, the predictive learning controller can improve the learning speed with less amount of information than the other learning algorithms require. And it is shown that the robustness of predictive learning controller is improved by using the prediction horizon. The predictive learning controller is also applied to find a reference trajectory for tuning of fuzzy rules, and to compensate the time-varying effect of batch reactors. Third, a hierarchical control structure is proposed as a generic form to be applied to processes of which the reaction heats are hard to obtain. The proposed control structure is a combination of three controllers: a nonlinear feedback with auxiliary variables, from an implementational point of view, a fuzzy controller can be a good solution; a feedforward controller; and an iterative learning controller to reject the disturbances occurring repetitively. An application of the proposed control structure to an industrial batch polymerization reactor shows that the batch cycle time is reduced by 16 \% and the standard deviations of product qualities are reduced by 40 \%. Fourth, to overcome the time-varying effect occurring in the industrial batch polymerization reactor mainly due to scaling of the reactor wall as batch operation proceeds, a feedback combined predictive learning control scheme is proposed. The jacket side of the reactor is modeled with the partial differential equation, and it is shown that the model represents the real situation well. It is also found that the proposed control scheme can handle the time-varying processes; the simulation study shows that control performance of the proposed scheme is still acceptable even when the overall heat transfer coefficient is decreased by 30\%.