Abstract
This paper presents a novel dynamic deep learning architecture integrated with Lyapunov control to address the timing latency and constraints of deep learning. The dynamic component permits the network depth to increase or decrease depending on the system complexity/nonlinearity evaluated through the parameterized complexity method. A correlation study between the parameter tuning effect on the error is also made thus causing a reduction in the deep learning time requirement and computational cost during the network training and retraining process. The control Lyapunov function is utilized as an input cost function to the DNN in order to determine the system stability. A relearning process is triggered to account for the introduction of disturbances or unknown model dynamics, therefore, eliminating the need for an observer-based approach. The introduction of the relearning process also allows the algorithm to be applicable to a wider array of cyber–physical systems (CPS). The intelligent controller autonomy is evaluated under different circumstances such as high frequency nonlinear reference, reference changes, or disturbance introduction. The dynamic deep learning algorithm is shown to be successful in adapting to such changes and reaching a safe solution to stabilize the system autonomously.
Highlights
Deep learning is a multi-layer technique for shallow machine learning that enables the neural network to learn complex nonlinear patterns
While deep-learning-based cyber–physical systems have a number of advantages, such as the ease with which complex patterns may be detected, ability to adapt and learn towards unknowns, and a higher degree of accuracy of predicting the outputs compared to shallow neural networks, but they present a number of downsides
Machine learning has been utilized in previous research along with usually what is described as a cost function, but one of the downsides found was the amount of time required for the deep neural network (DNN) to relearn during the process of updating the data set and the time requirement to come up with the appropriate solution [10,11]
Summary
Deep learning is a multi-layer technique for shallow machine learning that enables the neural network to learn complex nonlinear patterns. While deep-learning-based cyber–physical systems have a number of advantages, such as the ease with which complex patterns may be detected, ability to adapt and learn towards unknowns, and a higher degree of accuracy of predicting the outputs compared to shallow neural networks, but they present a number of downsides. Machine learning has been utilized in previous research along with usually what is described as a cost function, but one of the downsides found was the amount of time required for the DNN to relearn during the process of updating the data set and the time requirement to come up with the appropriate solution [10,11]
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
