Neural Network Based Feedback Scheduling of Multitasking Control Systems

To cope with resource constraints in multitasking control systems, feedback scheduling is emerging as an important technique for integrating control and scheduling. The ability of neural networks (NNs) as good and robust nonlinear function approximators, reducing the computation time as compared against algorithmic solutions, suggests applying them to the feedback scheduling problem. A novel, simple and intelligent feedback scheduler is presented using a feedforward NN. The algorithmic optimizer is utilized as a teacher to generate data samples for NN training. The role of the NN based feedback scheduler is to provide a good approximation to the optimal solution and online adjust the sampling period of each control task so that the overall system performance is maximized in the face of workload variations. The performance of the NN approach is evaluated through co-simulations of the scheduler, controllers and process dynamics.