Real-time optimization and computation for interconnected nonlinear systems using neural networks

We present a neural network approach for the real-time optimization and control of interconnected nonlinear systems in the presence of more general constraints, i.e. equality and inequality constraints, and bound-constrained variables. For the interconnected system with bound-constrained variables, we transform it into an equivalent formulation without bound constraints. With the help of auxiliary variables, the inequality constrained problem is reformulated as a problem with only equality constraints. Moreover, an electrocircuit is proposed for implementing the Lagrange neurons in the inequality constrained systems. Simulation studies show that this proposed method is satisfactory for the real-time optimization and control of large-scale systems.