Control of Complex Biological Systems Utilizing the Neural Network Predictor

Intelligent control of complex systems faces many challenges including difficulty in realizing the model of the system and the need to address uncertainties. Because a lot of data are collected in modern systems, a data-driven approach can be employed to design intelligent control algorithms. Specifically, machine learning can be used to take advantage of the available datasets and predict the behavior of the system for improved design and performance of the controller. For example, in this chapter, a time-shifted neural network predictor is integrated with a proportional–integral controller to compensate for performance errors associated with time lag and nonlinear absorption pattern of meal and insulin in closed-loop blood glucose control systems. Additional benefits of this approach include the mitigation of errors that may be associated with sensor drift and slow change in concentration of the interstitial fluid glucose measured by the continuous glucose monitors. Different control approaches and devices for blood glucose control were reviewed, and simulation studies were presented to show the effectiveness of a neural network integrated control approach.