Deep Neural Network Controller Research Articles

Event-Triggered Deep Learning Control of Quadrotors for Trajectory Tracking

This paper proposes an event-triggered deep learning control strategy to achieve real-time trajectory tracking control for quadrotors. In the training data collection phase, the event-triggered model predictive control (MPC) method is applied to the quadrotor in the simulation environment to generate training data. Then, a deep neural network (DNN) controller is trained to approximate the optimal control policy of the event-triggered MPC. To further save computing resources of on-board processor, the event-triggered mechanism is incorporated with the DNN controller, and the dual-mode approach is employed in it. Finally, simulation and experimental results show that the proposed controller can ensure almost similar trajectory tracking performance to the event-triggered MPC controller while requiring a lower control computation cost.

Implementation of Deep Learning-Based Bi-Directional DC-DC Converter for V2V and V2G Applications—An Experimental Investigation

Growth in renewable energy systems, direct current (DC) microgrids, and the adoption of electric vehicles (EVs) will substantially increase the demand for bi-directional converters. Precise control mechanisms are essential to ensure optimal performance and better efficiency of these converters. This paper proposes a deep neural network (DNN)-based controller designed to precisely control bi-directional converters for vehicle-to-vehicle (V2V) and vehicle-to-grid (V2G) applications. This control technique allows the converter to quickly attain new reference values, enhancing performance and efficiency by significantly reducing the overshoot duration. To train the DNN controller, large synthetic data are used by performing simulations for various sets of conditions, and the results are validated with a hardware setup. The real-time performance of the DNN controller is compared with a conventional proportional–integral (PI)-based controller through simulated results using MATLAB Simulink (version 2023a) and with a real-time setup. The converter attains a new reference of about 975 μs with the proposed control technique. In contrast, the PI controller takes about 220 ms, which shows that the proposed control technique is far better than the PI controller.

Deep learning-based model predictive control for real-time supply chain optimization

This paper presents a deep learning-based model predictive control (MPC) method for operational supply chain optimization in real time. The method follows an offline-online procedure. In the offline phase, the state-space model of a supply chain system is developed and the MPC problem for supply chain operation is formulated. Then, the MPC problem is solved for a set of initial states to obtain the corresponding optimal inputs. A deep neural network (DNN) is built and trained by using the optimal state-input pairs as training data to approximate the optimal MPC law. In the online phase, the DNN controller is employed to provide real-time decisions. In this paper, the MPC problem for supply chain operation is formulated as a mixed-integer linear program. A deep learning-based MPC method is proposed to accommodate time delays in the system. Moreover, a heuristic method is proposed for feasibility recovery with the binary decision variables taken into account. The training set for the DNN controller contains two subsets, one formed from MPC solutions corresponding to random initial states, and the other formed from optimal state-input pairs in closed-loop simulations. The deep learning-based MPC is validated via two case studies through closed-loop simulation. The first case study involves a linear MPC, and the second case study involves a more complicated mixed-integer linear MPC. Results show that deep learning-based MPC can achieve a high accuracy in approximating the MPC decisions and a significant reduction in the online computation time. Compared with MPC, the average performance loss of using deep learning-based MPC in the two cases is 0.43% and 1.8%, respectively.

Learning Deep Neural Network Controller for Path Following of Unicycle Robots

This letter investigates the scope of deep neural network (DNN) based controller in the path following task for unicycle mobile robots. A DNN-based controller is trained to follow paths with arbitrary curvature in two-dimensional space. The training process does not require initialization or supervision from any other known expert controller. Rather, the training of the DNN controller is guided by another predictive neural network that represents a path following error dynamics which is exponentially stable at the origin. The two DNNs are trained jointly in a simulated environment. The learned DNN controller is then employed as a standalone controller in a real unicycle robot for the tasks of following various linear and curved paths.

Cascade feedforward neural network and deep neural network controller on photovoltaic system with cascaded multilevel inverters: Comparison on standalone and grid integrated system

The introduction of a micro-grid-based power generation network will help to meet the demands of consumers while reducing environmental impact. Several industrialized and emerging countries allocate considerable resources to renewable energy-based power generation and invest significant sums of money in this area. This study examines the challenges involved with electricity generation through photovoltaic (PV) systems and the integration of the same with the grid to mitigate power quality issues and improve the power factor for various loading conditions. An innovative multilayer inverter for grid-connected PV systems has been developed to enhance the voltage profile and resulted in a drop in total harmonic distortion (THD). A cascade multilevel inverter (associated with a grid-integrated PV system and managed using multiple innovative artificial intelligence controllers) was developed in this research project. Various advanced intelligent controllers, such as cascade feedforward neural networks (CFFNN) and deep neural networks (DNN), have been analyzed under various operating situations and observed that the THD of voltage, current at the grid, and the load is less than 3 % as per the IEEE 519 standards along with this power factor is maintained nearly unity for the grid-connected system. The quality of power in terms of voltage, frequency, total harmonics distortion, and power factor are improved by using a novel deep neural network algorithm in a cascaded multilevel inverter and verified according to IEEE 1547 and IEEE 519 standards to determine the efficacy of the proposed system.

Using stochastic programming to train neural network approximation of nonlinear MPC laws

To facilitate the real-time implementation of nonlinear model predictive control (NMPC), this paper proposes a deep learning-based NMPC scheme, in which the NMPC law is approximated via a deep neural network (DNN). To optimize the DNN controller, a novel “optimize and train” architecture is designed, where the processes of data generation and neural network training are combined together to result in a single large-scale stochastic optimization problem. Unlike the conventional “optimize then train” approach, our proposed one directly optimizes the closed-loop performance of the DNN controller over a finite horizon for a number of initial states. The important features of our proposed scheme are that it can deal with set-valued optimal MPC input, and a probabilistic guarantee of constraint satisfaction can be concluded for the closed-loop system without simulating the DNN controller. With our proposed scheme, an increased number of training scenarios leads to improved constraint satisfaction of the derived DNN controller, which is not necessarily true for the “optimize then train” approach. Statistical approaches for validating closed-loop control performance are also discussed. Furthermore, computational methods are introduced to efficiently solve the resulting stochastic optimization problem. The effectiveness of the proposed scheme is extensively illustrated with several numerical simulations. Compared with the conventional “optimize then train” approach, our proposed approach exhibits better closed-loop constraint satisfaction for all considered case studies.

A guidance method for coplanar orbital interception based on reinforcement learning

This paper investigates the guidance method based on reinforcement learning (RL) for the coplanar orbital interception in a continuous low-thrust scenario. The problem is formulated into a Markov decision process (MDP) model, then a well-designed RL algorithm, experience based deep deterministic policy gradient (EBDDPG), is proposed to solve it. By taking the advantage of prior information generated through the optimal control model, the proposed algorithm not only resolves the convergence problem of the common RL algorithm, but also successfully trains an efficient deep neural network (DNN) controller for the chaser spacecraft to generate the control sequence. Numerical simulation results show that the proposed algorithm is feasible and the trained DNN controller significantly improves the efficiency over traditional optimization methods by roughly two orders of magnitude.

A Study of Deep Neural Network Controller-Based Power Quality Improvement of Hybrid PV/Wind Systems by Using Smart Inverter

Presently, climate change and global warming are the most uncontrolled global challenges due to the extensive fossil fuel usage for power generation and transportation. Nowadays, most of the developed countries are concentrating on developing alternative resources; consequently, they did huge investments in research and development. In general, alternative energy resources including hydropower, solar power, and wind energy are not harmful to nature. Today, solar power and wind power are very popular alternative energy sources due to their enormous availability in nature. In this paper, the photovoltaic cell and wind energy systems are investigated under various weather conditions. Based on the findings, we developed an advanced intelligent controller system that tracks the maximum power point. The MPPT controller is a must for the renewable energy sources due to unpredictable weather conditions. The main objective of this paper is to propose a new algorithm that is based on deep neural network (DNN) and maximum power point tracking (MPPT), which was simulated in a MATLAB environment for photovoltaic (PV) and wind-based power generation systems. The development of an advanced DNN controller that improves the power quality and reduces THD value for the microgrid integration of hybrid PV/wind energy system was performed. The MATLAB simulation tool has been used to develop the proposed system and tested its performance in different operating situations. Finally, we analyzed the simulation results applying the IEEE 1547 standard.

Deep Neural Network Approximation of Nonlinear Model Predictive Control

This paper focuses on developing effective computational methods to enable the real-time application of model predictive control (MPC) for nonlinear systems. To achieve this goal, we follow the idea of approximating the MPC control law with a Deep Neural Network (DNN). To train the deep neural network offline, we propose a new “optimize and train” method that combines the steps of data generation and neural network training into a single high-dimensional stochastic optimization problem. This approach directly optimizes the closed loop performance of the DNN controller over a finite horizon for a number of initial states. The large-scale optimization problem can be solved efficiently using parallel computing techniques. The benefits of this approach over the conventional “optimize then train” protocol is illustrated through numerical results.

A Deep Learning Strategy For On-Orbit Servicing Via Space Robotic Manipulator

Autonomous robotic systems are currently being addressed as a critical element in the development of present and future on-orbit operations. Modern missions are calling for systems capable of reproducing human’s decision-making process, thus enhancing their performance. Generally, space manipulators are mounted on a floating spacecraft in a microgravity environment, consequently leading to a mutual influence between the robotic arms and the platform dynamics, thus making the motion planning and control design more challenging than those of terrestrial robots. Another aspect to be considered is that space robots are designed as lightweight systems, resulting in a significant dynamic coupling between their rigid motion and structural elasticity. These effects involve critical issues in modelling their dynamics and designing a suitable controller. In this context, Deep Neural Network (DNN) architectures and the related Deep Learning (DL) techniques have widely proved to have powerful capability in solving data-driven nonlinear modelling problems and they can hence represent a viable solution for space activities. The present paper deals with the design of a DNN controller for a space manipulator system, which has to follow a specific path for a typical on-orbit servicing mission. The goal is to provide proper control inputs autonomously adapting to the given desired trajectory. Structural flexibility and joint friction features are implemented in the dynamic model as well.

Deep learning framework for controlling an active suspension system

In this paper, a feed-forward deep neural network (DNN) and automated search method for optimum network structure are developed to control an active suspension system (ASS). The network was trained through supervised learning using the backpropagation algorithm. The training data were generated from an optimal proportional–integral–derivative controller tuned based on a full state feedback optimal controller. The trained network was implemented in an ASS test rig for a quarter-car model and was initially tested in simulation under parameter uncertainties. Experimental results showed that the developed DNN controller outperforms the optimal controller under uncertainties in terms of reducing the sprung mass acceleration and actuator energy consumption, with a 4% and 14% reduction, respectively.

Handover Control in Wireless Systems via Asynchronous Multiuser Deep Reinforcement Learning

In this paper, we propose a two-layer framework to learn the optimal handover (HO) controllers in possibly large-scale wireless systems supporting mobile Internet-of-Things users or traditional cellular users, where the user mobility patterns could be heterogeneous. In particular, our proposed framework first partitions the user equipments (UEs) with different mobility patterns into clusters, where the mobility patterns are similar in the same cluster. Then, within each cluster, an asynchronous multiuser deep reinforcement learning (RL) scheme is developed to control the HO processes across the UEs in each cluster, in the goal of lowering the HO rate while ensuring certain system throughput. In this scheme, we use a deep neural network (DNN) as an HO controller learned by each UE via RL in a collaborative fashion. Moreover, we use supervised learning in initializing the DNN controller before the execution of RL to exploit what we already know with traditional HO schemes and to mitigate the negative effects of random exploration at the initial stage. Furthermore, we show that the adopted global-parameter-based asynchronous framework enables us to train faster with more UEs, which could nicely address the scalability issue to support large systems. Finally, simulation results demonstrate that the proposed framework can achieve better performance than the state-of-art online schemes, in terms of HO rates.

A Deep Learning Architecture for Predictive Control

Model predictive control (MPC) is a popular control strategy that computes control actions by solving an optimization problem in real-time. Uncertainty and nonlinearity of a process, and the non-convexity of the resulting optimization problem can make online implementation of MPC nontrivial. Consequently, MPC is most often used in processes where the time constants are large and/or high-performance computing support is available. We propose a deep neural network (DNN) controller architecture to reduce the computational cost of implementing an MPC. This is done by training a DNN controller on simulated input-output data from a well-designed MPC. The online implementation of a DNN controller does not require solving an optimization problem. Once the DNN is trained, the MPC is fully replaced with the DNN controller. The benefits of this approach are illustrated through a simulated example.