Dynamic Surrogate Modeling for Continuous Processes Control Applications
Résumé
With the increasing amount of data to quickly process, surrogate modeling has become a topic of major interest in process engineering during the last decades. Recently, black-box models have gained renewed interest as effective tools to reduce the computational effort and to enhance the solution of optimization algorithms. Among the surrogate model applications, scheduling and control problems are getting particular attention in the last years. However, when dealing with dynamics, the majority of literature works concern batch processes to be modeled within the corresponding residence time interval. With the purpose to extend the established procedures to continuous operations, in this research work the surrogate modeling procedure of a dynamic non-isothermal CSTR reactor is discussed and analyzed in detail. The modeling phase was carried out for the temperature and concentration with different sampling size and techniques by means of the software ALAMO®. After the open-loop model was built and validated, the related closed loop configuration has been tested by means of a conventional PID controller implementation. The controller has been properly tuned on both models in order to compare the performances. The obtained trends show good agreement with those of the rigorous model both for open loop and closed loop performances. More importantly, the implemented surrogate model requires reduced calculation time thanks to the explicit input-output variable correlations. In conclusion, the proposed dynamic surrogate modeling approach for process control applications has proved to be effective and provided reliable results. Moreover, the computational time was reduced by an order of magnitude and the best compromise between the performances and accuracy can be detected.