A robust framework with statistical learning method and evolutionary improvement algorithm for process real-time optimization

This study proposes an effective framework for process real-time optimization and data-driven modeling method. The proposed RTO framework with evolutionary improvement algorithm does not wait for the steady-state and it corrects the set-point continuously through the similar way which genetic algorithm exploit to find optimal points. It can deal with higher frequency disturbances and is less influenced by control system performance. Moreover, it is able to address the convergence to suboptimal. Also, this study proposes statistical learning model (modified support vector machine) that is used in RTO framework. It is able to handle highly-nonlinearity and carry out parameter tuning easily. The performance of proposed method was successfully illustrated by means of RTO example.