A Fast Model Predictive Control Method for Air Separation Plant Based on FPAA Simulation Neural Network

Abstract

The invention discloses a rapid model prediction control method for an air separation unit based on an FPAA simulation neural network. The method comprises two parts: offline calculation and online calculation. The offline calculation comprises the calculation of an MPC control parameter and the construction of an analog circuit QP solver; The online calculation comprises state updating, unconstrained optimization, calculation of translation transformation and scale transformation parameters, and QP solving of the analog circuit. The method employs a continuous neural network for solving a QPproblem, and has the natural parallelism. The method solves a problem that the analog circuit achieves the signal restriction of a continuous neural network through the translation transformation andscale transformation. On the basis, the circuit employs an FPAA for designing the analog circuit, thereby achieving the rapid solving of the QP in the MPC. Compared with an existing technology, the method effectively irons out the defects of a discrete neural network, and solves the problems in a better way that a conventional numerical method is low in MPC solving speed and is poorer in real-timeperformance.

Description

technical field [0001] The present invention relates to the fast model predictive control (Fast MPC) field of air separation unit, particularly relate to a kind of fast model predictive control framework based on FPAA simulation neural network, it is characterized in that adopting FPAA simulation neural network to solve the QP problem in MPC, has High real-time performance. Background technique [0002] Model predictive control (Model Predictive Control, referred to as MPC) has been widely used in petroleum, chemical and other process fields due to its excellent constraint optimization control ability in complex multivariable systems. In essence, the constraint optimization control capability of MPC is mainly generated from the online solution of the quadratic programming (Quadratic Programming, QP) problem with constraints. Although the traditional numerical solution algorithm of quadratic programming has a wide range of applications, most of them involve operations such a...

AI Technical Summary

Problems solved by technology

Although the traditional numerical solution algorithm of quadratic programming has a wide range of applications, most of them involve operations such as matrix inversion and decomposition, so the real-time performance is low, so the main application of MPC is still limited to slow systems with long control cycles such as process industries

In the current simulation optimization research, most of the simulation circuit parameters cannot be updated online

Images