13 April 2023 in IEEE Transactions on Neural Networks and Learning Systems
This article proposes a deep learning (DL)-based control algorithm—DL velocity-based model predictive control (VMPC)—for reducing traffic congestion with slowly time-varying traffic signal controls. This control algorithm consists of system identification using DL and traffic signal control using VMPC. For the training process of DL, we established a modeling error entropy loss as the criteria inspired by the theory of stochastic distribution control (SDC) originated by the fourth author. Simulation results show that the proposed algorithm can reduce traffic congestion with a slowly varying traffic signal control input. Results of an ablation study demonstrate that this algorithm compares favorably to other model-based controllers in terms of prediction error, signal varying speed, and control effectiveness.
7 December 2023 in IEEE Access
In recent years, deep learning has been widely applied to learning nonlinear dynamic models for the development of a digital twin system. However, most traditional deep learning frameworks, such as recurrent neural networks, convolutional neural networks, and multilayer perceptrons, find it difficult to learn continuous-time and nonlinear system models. To address this challenge, in this paper, a novel deep learning method called neural controlled differential equation has been proposed to model the unknown nonlinear dynamics of controlled continuous-time systems seen in Francis hydroturbines of hydropower systems. Following the development of discretized-model structures for the system using the first principles, a detailed learning algorithm is formulated that is integrated with the physical model of the hydroturbine. As a result, a hybrid modeling with effective learning capability is obtained. To test the effectiveness of the proposed learning algorithm, a set of operational data has been collected and used to train the nonlinear dynamics of the Francis hydroturbine, where the learning results of the two nonlinear dynamics, namely the mechanical torque and water flow dynamics, using the real data have indicated that the proposed method can accurately learn these unknown nonlinear dynamics in an online, adaptive way. Moreover, to address the overfitting problem that appears during the online training phase, we propose to apply a meta-learning technique to pre-train a meta-initial value for each parameter of the proposed neural controlled differential equations. It has been shown that the use of the meta-learning technique can reduce the prediction mean square error significantly by more than 60%.
11 June 2025 in IEEE Access
This paper introduces a gain-scheduling PID controller design strategy for hydroturbine frequency control mode. This scheme first uses real data to learn the nonlinear dynamics of the hydroturbine using neural controlled differential equations and then perturbs the obtained nonlinear system at different equilibrium points, based on which a static output feedback adaptive dynamic programming algorithm is then used to optimize the PID gains for each equilibrium point. Moreover, a continuous-time version of stochastic distribution control is proposed to further fine-tune the optimized PID gains. Finally, the controller is obtained by implementing linear interpolation between the optimized PID control gains. The simulation results show that the proposed gain-scheduling PID controller can control a larger range of operation points compared with the given fixed PID controller and the baseline method. Compared with the given fixed PID controller, the proposed gain-scheduling PID controller can regulate hydroturbine frequency against disturbances induced by power-load variation with over 50% less overshoot for some operation points.
23 January 2026 in Transportation Research Part B: Methodological
11 November 2025 in IEEE Transactions on Networking
Routing configurations of a network should constantly adapt to traffic variations to achieve good network performance. Adaptive routing faces two main challenges: 1) how to accurately measure/estimate time-varying traffic matrices? 2) how to control the network and application performance degradation caused by frequent route changes? In this paper, we develop a novel data-enabled predictive traffic engineering (DeeP-TE) algorithm that minimizes the network congestion by gracefully adapting routing configurations over time. Our control algorithm can generate routing updates directly from the historical routing data and the corresponding link rate data, without direct traffic matrix measurement or estimation. Numerical experiments on real network topologies with real traffic matrices demonstrate that the proposed DeeP-TE routing adaptation algorithm can achieve close-to-optimal control effectiveness with significantly lower routing variations than the baseline methods.