Scalable Non-dimensional Model Predictive Control of Liquid Level in Generally Shaped Tanks Using RBF Neural Network

This paper focuses on developing and analyzing a concept of a fully scalable control method applicable to highly nonlinear systems with dynamics varying over the whole working area. The approach is demonstrated on control of liquid level in non-trivial shaped tanks. Non-dimensionalised quantities were used for the development of general geometric model systems of the liquid accumulation in the tanks. Then, training sets were obtained from simulations of the model systems and used for training radial basis function neural network (RBFNN) models. These RBFNNs were implemented in controllers using model predictive control (MPC) method. Both the models and controllers are scalable and applicable in industry or nature. A tentative set of conditions and rules was defined to transfer the solution to practical situations.

Automated summary

This paper focuses on developing and analyzing a fully scalable control method for highly nonlinear systems with dynamically changing dynamics across the entire working area. The approach is demonstrated through the control of liquid levels in non-trivial shaped tanks. Non-dimensionalized quantities are utilized to develop general geometric model systems for liquid accumulation in the tanks. Radial basis function neural network (RBFNN) models are trained using simulation data from the model systems and implemented in controllers using model predictive control (MPC) method. These scalable models and controllers are applicable in both industrial and natural environments, with a tentative set of conditions and rules defined for practical implementation.