Network Controller Research Articles

Control-error-based output-feedback adaptive decentralized neural network controller for interconnected uncertain strict-feedback nonlinear systems with input saturation

In this paper, a control-error-based decentralized neural network (NN) direct adaptive controller is presented for uncertain interconnected nonlinear systems, in strict-feedback form, subject to input saturation and external disturbances with unavailable states for measurement. Different from the existing results in the literature, the proposed approach is based on the control error instead of the tracking error resulting in a separation-like principle. Furthermore, the explosion of complexity due to back-stepping recursive design is completely avoided along with discarding all restrictive assumptions imposed on the unmatched interconnections. Actually, NNs are used to approximate the unknown ideal control laws, and auxiliary control terms are appended to deal with approximation errors and enhance the stability of the closed-loop system. Besides, fuzzy inference systems are introduced to estimate the unknown control errors, leading to simplified derivation of adaptive laws. Thanks to the strictly positive real (SPR) property, the tracking errors are proved to converge asymptotically to zero using Lyapunov theory, which is superior to bounded stability results usually found in the literature. Simulation results show the effectiveness of the proposed approach.

Mitigation of Grid Current Harmonics by ABC- ANN based Shunt Active Power Filter

Harmonics are being introduced into power system networks as a result of the increasing use of nonlinear devices. These harmonics cause distortion of current and voltage signals, which in turn causes damage to power distribution systems. As a result, the suppression of harmonics is of extreme significance in power systems. This paper proposes Shunt Active Power Filters (SAPF) based on neural network algorithms like Artificial Neural Network (ANN) as a feasible approach to mitigating harmonic distortion and raising power quality in electrical distribution systems. This research shows that using shunt active power filters (SAPF), which use the Artificial Bee Colony Optimized Artificial Neural Network Controller (ABC-ANN), is an efficient method for enhancing power quality and minimizing harmonic distortion in distribution systems. The ABC-ANN algorithms have been produced for SAPF with the goal of improving system performance by reducing Grid current Harmonics .In the first stage of this work, PSO Technique is used to tune a standard PI controller to its optimum gain values (Ki ,Kp). Then, these target and input data of PSO tuned PI controller will be supplying inputs to the ANN controller. To find the optimal weight and bias values, this ANN controller has been tuned with the help of the ABC algorithm. Using MATLAB/SIMULINK software, we compare the performance of the proposed algorithm to that of other optimization algorithm like Differential Evaluation (DE) algorithm, as well as the PSO tuned PI controller and traditional PI controller. The findings from the simulation suggest that a SAPF utilizing an ABC trained ANN controller could improve THD in the supplying current while maintaining harmonics within IEEE-519 accepting levels.

Leader-Follower UAV formation flight control based on feature modelling

To solve the problems of backstepping error and poor dynamic tracking approach rate in traditional PID neural network control in UAV formation flight control, a Leader-Follower UAV formation flight control method based on feature modelling is proposed,and the pose relationship model between virtual follower and pilot is established by trajectory tracking and pose dynamic fitting. The pose distribution of thefollower is analyzed in the ground coordinate system, and the parameter information of linear velocity and angular velocity control of UAV is obtained, and the backstepping sliding mode formation controller is formed. The variable structure PID neural network controller is used to design the flight control law of UAV formation, and the fast piecewise power approaching factor is introduced into the PID controller to eliminate the chattering of sliding mode control. The simulation results show that this method can ensure the rapidity of UAV formation flight control also show strong anti-jamming ability. Due to the fast piecewise power approach rate, the UAVs can complete the UAV formation reorganization under disturbance and buffeting in a short time, and the trajectory tracking error approaches zero, and it has good anti-buffeting ability.

A Novel Actor—Critic Motor Reinforcement Learning for Continuum Soft Robots

Reinforcement learning (RL) is explored for motor control of a novel pneumatic-driven soft robot modeled after continuum media with a varying density. This model complies with closed-form Lagrangian dynamics, which fulfills the fundamental structural property of passivity, among others. Then, the question arises of how to synthesize a passivity-based RL model to control the unknown continuum soft robot dynamics to exploit its input–output energy properties advantageously throughout a reward-based neural network controller. Thus, we propose a continuous-time Actor–Critic scheme for tracking tasks of the continuum 3D soft robot subject to Lipschitz disturbances. A reward-based temporal difference leads to learning with a novel discontinuous adaptive mechanism of Critic neural weights. Finally, the reward and integral of the Bellman error approximation reinforce the adaptive mechanism of Actor neural weights. Closed-loop stability is guaranteed in the sense of Lyapunov, which leads to local exponential convergence of tracking errors based on integral sliding modes. Notably, it is assumed that dynamics are unknown, yet the control is continuous and robust. A representative simulation study shows the effectiveness of our proposal for tracking tasks.

Radial basis function neural network controller for power control of molten salt breeder reactor of nuclear power plant

The molten salt breeder reactor (MSBR) basically uses molten salt as fuel. Due to molten salt, it is difficult to control power of the MSBR core. To solve this problem, a neural network control scheme is adopted for the same system. Firstly, a nonlinear MSBR model is taken and after linearizing a single input single output (SISO) system is constructed out of it. Radial basis function neural network (RBFNN) has strong tolerance to input noise, online learning ability and good generalization. Therefore, RBFNN controller is used for controlling the core power of the MSBR. The conventional RBFNN dominates in the neural active region. Compact RBF networks has been designed by using newly developed algorithm. By using Lyapunov stability criterion, it has proved that the variation of core power is bounded and it may be comparable to any traditional neural networks or fuzzy interference system. Results of the study reflect the efficiency of RBFNN controller for controlling power of MSBR core.

Dual Neural Network Control of a Hybrid Functional Electrical Stimulation Cycling System

Abstract Hybrid functional electrical stimulation (FES) cycling is a method to rehabilitate people with neurological conditions when they are not in and of themselves capable of fully controlling their extremities. To ensure smooth cycling and adequate stimulation to accomplish the rehabilitation task, admittance control is applied between the human and the robotic cycle. The cycle motor is actuated by a dual neural network control structure with an additional robust element tracking the admittance trajectory, while muscles are stimulated with a simple saturated robust controller. The dual neural network structure allows adaptation to separable functions of the dynamic system, in addition to shared adaptation through the admittance filter. A Lyapunov analysis shows that the admittance tracking controller is globally exponentially stable. A passivity analysis shows that the admittance system and cadence tracking error are output strictly passive. A combined analysis shows that the total system is passive. Experiments are performed on eight participants without neurological conditions, on 12 differing protocols including a robust controller for comparison, the addition of noise, and the addition or lack of stimulation. One participant with a neurological condition was evaluated on three different protocols, including a robust controller, a neural network controller, and a game-like mode where the participant was asked to track the trajectory as it appeared on a screen. Statistical analysis of the experiments show that the standard deviation of the tracking error is significantly improved with the adaptive dual neural network control addition when compared to the robust controller, in some instances reducing the magnitude by half.

Neural-Network-Based Time Control for Microwave Oven Heating of Food Products Distributed by a Solar-Powered Vending Machine with Energy Management Considerations

This article presents novel research on the utilization of a neural-network-based time control system for microwave oven heating of food items within a solar-powered vending machine. The research aims to explore the control of heating time for various food products, considering multiple variables. The neural network controller is calibrated through extensive experimentation, allowing it to accurately predict optimal heating times based on input parameters such as food type, weight, initial temperature, water content, and desired doneness level. The results demonstrate that the neural-network-controlled microwave oven achieves precise and desirable heating durations, mitigating the risk of overheating and ensuring superior food quality and taste. Moreover, the solar-powered vending machine showcases a commitment to sustainable energy sources, effectively reducing dependence on non-renewable energy and minimizing greenhouse gas emissions. To maintain food quality and freshness, a food refrigeration unit is integrated into the vending machine, employing load-balancing technology to control the refrigeration chamber’s temperature effectively. Energy efficiency is prioritized in both the refrigeration unit and the microwave oven through intelligent algorithms and system optimization. The combination of a neural-network-controlled microwave oven, a solar-powered vending machine, and a food refrigeration unit introduces a novel and sustainable approach to food preparation and energy management.

Learning a simulation-based visual policy for real-world peg in unseen holes.

This paper proposes a learning-based visual peg-in-hole that enables training with several shapes in simulation and adapting to arbitrary unseen shapes in the real world with minimal sim-to-real cost. The core idea is to decouple the generalization of the sensory-motor policy from the design of a fast-adaptable perception module and a simulated generic policy module. The framework consists of a segmentation network (SN), a virtual sensor network (VSN), and a controller network (CN). Concretely, the VSN is trained to measure the pose of the unseen shape from a segmented image. After that, given the shape-agnostic pose measurement, the CN is trained to achieve a generic peg-in-hole. Finally, when applying to real unseen holes, we only have to fine-tune the SN required by the simulated VSN + CN. To further minimize the transfer cost, we propose to automatically collect and annotate the data for the SN after one-minute human teaching. Simulated and real-world results are presented under the configuration of eye-to/in-hand. An electric vehicle charging system with the proposed policy inside achieves a 10/10 success rate in 2-3s, using only hundreds of auto-labeled samples for the SN transfer.

H∞ Bipartite Synchronization Control of Markov Jump Cooperation-Competition Networks With Reaction-Diffusions.

This article is concerned with the bipartite synchronization problem of coupled switching neural networks with cooperative-competitive interactions and reaction-diffusion terms. Different from the existing literature, the networked systems under investigation possess the relationship of cooperation and competition among nodes. Notably, the switching topology is described by a signed graph subject to the Markov jump process with the coexistence of positive and negative interaction weights. Specifically, a positive weight indicates an alliance relationship between two nodes and a negative one shows an adversary relationship. This article aims to design a bipartite synchronization controller for the aforementioned networks with the switching topology such that a prescribed H∞ bipartite synchronization is satisfied. Then, some sufficient criteria to ensure the stochastic stability of bipartite synchronization error systems are established in view of an appropriate Lyapunov function. Finally, two simulation examples are presented to verify the validity of the proposed bipartite synchronization control method.

Novel General Regression Neural Networks for Improving Control Accuracy of Nonlinear MIMO Discrete-Time Systems.

In this article, a novel version of the general regression neural network (Imp_GRNN) is developed to control a class of multiinput and multioutput (MIMO) nonlinear discrete-time (DT) systems. The improvements retain the features of the original GRNN along with a significant improvement of the control accuracy. The enhancements include developing a method to set the input-hidden weights of GRNN using the inputs recursive statistical means, introducing a new output layer and adaptable forward weighted connections from the inputs to the new layer, and suggesting an interval-type smoothing parameter to eradicate the need for selecting the parameter beforehand or adapting it online. Also, controller stability is studied using Lyapunov's method for DT systems. The controller performance is tested with different simulation examples and compared with the original GRNN to verify its superiority over it. Also, Imp_GRNN performance is compared with an adaptive radial basis function network controller, an adaptive feedforward neural-network (NN) controller, and a proportional-integral-derivative (PID) controller, where it demonstrated higher accuracy in comparison with them. In comparison with the formerly proposed control methods for MIMO DT systems, our controller is capable of producing high control accuracy while it is model free, does not require complex mathematics, has low computational complexity, and can be utilized for a wide range of DT dynamic systems. Also, it is one of the few methods that aims to improve the control system accuracy by improving the NN structure.

SAPF based on ANN-Second Order Generalized Integrator for Harmonic Absorption

Abstract: The purpose of this study is to provide a photovoltaic (PV)-based generation system that is coupled with a shunt active power filter (SAPF) in order to effectively compensate for reactive power and mitigate harmonics. The Solar Active Power Filter is made up of a DC link capacitor, a voltage source inverter (VSI), and a solar production system. With the assistance of an active power filter, the current harmonics that are brought on by nonlinear loads are capable of being significantly decreased. The artificial neural network is presented as a solution for both the generation of the reference current and the regulation of SAPF. Calculating the reference source current for SAPF requires the use of the Second Order Generalized Integrator (SOGI), which is coupled with an Artificial Neural Network (ANN) controller. ANN is also notable for its high compatibility with digital implementation, control performance, and blazingly quick dynamic reaction times. The created controller is validated with the assistance of MATLAB simulations in order to demonstrate the efficacy and superior concert of the proposed methodology

Neural Network Learning Algorithms for High-Precision Position Control and Drift Attenuation in Robotic Manipulators

In this paper, different learning methods based on Artificial Neural Networks (ANNs) are examined to replace the default speed controller for high-precision position control and drift attenuation in robotic manipulators. ANN learning methods including Levenberg–Marquardt and Bayesian Regression are implemented and compared using a UR5 robot with six degrees of freedom to improve trajectory tracking and minimize position error. Extensive simulation and experimental tests on the identification and control of the robot by means of the neural network controllers yield comparable results with respect to the classical controller, showing the feasibility of the proposed approach.

Towards Future-Based Explanations for Deep RL Network Controllers

Lack of explainability is hindering the practical adoption of high-performance Deep Reinforcement Learning (DRL) controllers. Prior work focused on explaining the controller by identifying salient features of the controller's input. However, these feature-based methods focus solely on inputs and do not fully explain the controller's policy. In this paper, we put forward future-based explainers as an essential tool for providing insights into the controller's decision-making process and, thereby, facilitating the practical deployment of DRL controllers. We highlight two applications of futurebased explainers in the networking domain: online safety assurance and guided controller design. Finally, we provide a roadmap for the practical development and deployment of future-based explainers for DRL network controllers.

Memory capacity of recurrent neural networks with matrix representation

It is well known that canonical recurrent neural networks (RNNs) face limitations in learning long-term dependencies, which have been addressed by memory structures in long short-term memory (LSTM) networks. Neural Turing machines (NTMs) are a variant of RNNs that implement the notion of programmable memory with neural network controllers that can learn simple algorithmic tasks. Matrix neural networks, on the other hand, feature matrix representations which inherently has the potential to preserve the spatial structure of data when compared to canonical neural networks that use only vector-based representation. One may then argue that neural networks with matrix representations may have the potential to provide better memory capacity. In this paper, we define and study a probabilistic notion of memory capacity based on Fisher information for matrix-based RNNs. We find bounds on memory capacity for such networks under various hypotheses and compare them with their conventional (vector) counterparts. In particular, we show that the memory capacity of such networks is bounded by N2, for N×N state matrix which generalizes and provides similar results for vector networks. We also show and analyze the increase in memory capacity for such networks which is introduced when one exhibits an external state memory, such as NTMs. Consequently, we construct NTMs with RNN controllers with matrix-based representation of external memory, leading us to introduce Matrix NTMs. We demonstrate the performance of this class of memory networks under certain algorithmic learning tasks such as copying and recall and compare it with Matrix RNNs. We find an improvement in the performance of Matrix NTMs by the addition of external memory, in comparison to Matrix RNNs.

Dual Event-Triggered Controller Co-Design for Networked Control Systems with Network-Induced Delays

To address the presence of network-induced delays in networked control systems (NCSs), a dual event-triggered mechanism (DETM) is used to investigate the problem of reducing network delays and controller co-design. Firstly, the DETM of the sensor–controller (SC) and the controller–actuator (CA) is adopted. By determining whether the sampled data meet the event-triggered threshold conditions for network transmission, we effectively reduce the sampled data transmitted over the network, which can reduce a network delay by reducing occupation of the network resources. Secondly, a dual event-triggered NCS model with a network-induced delay is developed, and a Lyapunov function including a DETM and network-induced delay is chosen. The functional upper limit of the Lyapunov function is estimated by combining the Wirtinger’s-based integral inequality with the reciprocally convex approach. This results in a stability criterion for systems with low conservativeness and a controller co-design method for a DETM. Finally, the availability of this method was verified through a numerical example and case study.

Salp Swarm Optimized Hybrid Elman Recurrent Neural Network (SSO-ERNN) based MPPT Controller for Solar PV

Renewable energy technologies provide clean and abundant energy that can be self-renewed from natural sources; more support from the public to replace fossil fuels with various renewable energy sources to protect the environment. Although solar energy has less impact on the environment than other renewable sources, the output efficiency is lower due to the different weather conditions. So to overcome that, the MPPT controller is used for tracking peak power and better efficiency. Some conventional methods in MPPT controllers provide less tracking efficiency, and steady-state oscillations occur in maximum power tracking due to the sudden variations in solar irradiance. Thus, in this work salp swarm optimized (SSO) based Elman recurrent neural network (ERNN) controller is proposed to track the maximum power form PV with high efficiency. The weight parameter of ERNN layer is optimized with the help of SSO, which solve the complex problems and give maximum efficiency. The proposed method is performed in MATLAB/Simulink environment, which differs from existing plans and gives a better output efficiency. Using this proposed controller, the system can achieve high tracking efficiency of 99.74% compared to conventional processes.

Developing an artificial neural network controller for accelerating the hot deformation of the titanium aluminide TNM-B1 using reinforcement learning and finite element simulations

This work presents a framework for interfacing a reinforcement learning algorithm with a finite element model in order to develop an artificial neural network controller. The goal of the controller is accelerating the hot compression process of the titanium aluminide TNM-B1. The reinforcement learning algorithm interacts with the finite element model by exploring different die velocities and receiving input measurements (the velocity, displacement and force of the die) while collecting rewards if a constant stress state in the workpiece is achieved. Synthetic stochastic material behavior was used to simulate the observed variations in deformation behavior of TNM-B1. The same reinforcement learning setup and reward function was able to adapt to two example finite element environments; the compression of a simple cylinder workpiece between flat dies and the compression of a more complex bone workpiece between flat dies. The performance of the controller for the bone compression environment was comparatively reduced and less consistent. In addition, training times and training instability were significantly increased. Furthermore, the results suggest that the framework can be used as a tool to find process optimizations or alternative process routes. This work demonstrates the concept and provides the groundwork and fundamentals for transferring the method to a physical setup.

Control for Bioethanol Production in a Pressure Swing Adsorption Process Using an Artificial Neural Network

This paper introduces a new approach to controlling Pressure Swing Adsorption (PSA) using a neural network controller based on a Model Predictive Control (MPC) process. We use a Hammerstein–Wiener (HW) model representing the real PSA process data. Then, we design an MPC-controlled model based on the HW model to maintain the bioethanol purity near 99% molar fraction. This work proposes an Artificial Neural Network (ANN) that captures the dynamics of the PSA model controlled by the MPC strategy. Both controllers are validated using the HW model of the PSA process, showing great performance and robustness against disturbances. The results show that we can follow the desired trajectory and attenuate disturbances, achieving the purity of bioethanol at a molar fraction value of 0.99 using the ANN based on the MPC strategy with 94% of fit in the control signal and a 97% fit in the purity signal, so we can conclude that our ANN can be used to attenuate disturbances and maintain purity in the PSA process.

Adaptive neural network control of multiple‐sectioned flexible riser with time‐varying output constraint and input nonlinearity

AbstractIn this paper, an adaptive neural network controller is proposed for vibration suppression of a multisectional riser system with unknown boundary disturbance, time‐varying asymmetric output constraint, and input nonlinearity. The considered riser system is composed of a continuous connection of several different pipes, and its dynamic models are represented by a set of multiple continuously connected partial differential equations (PDEs) and an ordinary differential equation (ODE) at the top boundary. Considering input nonlinearity, external disturbance, and system uncertainty, radial basis function (RBF) neural networks are adopted to eliminate the effect of these uncertain terms. Besides, a barrier Lyapunov function is employed to guarantee the restrictions. With the proposed boundary control, the stability of the closed‐loop system is proved and simulations are given to illustrate the well performance of the proposed control strategy.

Reaching the limit in autonomous racing: Optimal control versus reinforcement learning.

A central question in robotics is how to design a control system for an agile mobile robot. This paper studies this question systematically, focusing on a challenging setting: autonomous drone racing. We show that a neural network controller trained with reinforcement learning (RL) outperformed optimal control (OC) methods in this setting. We then investigated which fundamental factors have contributed to the success of RL or have limited OC. Our study indicates that the fundamental advantage of RL over OC is not that it optimizes its objective better but that it optimizes a better objective. OC decomposes the problem into planning and control with an explicit intermediate representation, such as a trajectory, that serves as an interface. This decomposition limits the range of behaviors that can be expressed by the controller, leading to inferior control performance when facing unmodeled effects. In contrast, RL can directly optimize a task-level objective and can leverage domain randomization to cope with model uncertainty, allowing the discovery of more robust control responses. Our findings allowed us to push an agile drone to its maximum performance, achieving a peak acceleration greater than 12 times the gravitational acceleration and a peak velocity of 108 kilometers per hour. Our policy achieved superhuman control within minutes of training on a standard workstation. This work presents a milestone in agile robotics and sheds light on the role of RL and OC in robot control.