Feedforward Neural Network Controller Research Articles

Fuzzy inference system enabled neural network feedforward compensation for position leap control of DC servo motor

To improve dynamic performance and steady-state accuracy of position leap control of the direct current (DC) servo motor, a fuzzy inference system (FIS) enabled artificial neural network (ANN) feedforward compensation control method is proposed in this study. In the method, a proportional-integral-derivative (PID) controller is used to generate the baseline control law. Then, an ANN identifier is constructed to online learn the reverse model of the DC servo motor system. Meanwhile, the learned parameters are passed in real-time to an ANN compensator to provide feedforward compensation control law accurately. Next, according to system tracking error and network modeling error, an FIS decider consisting of an FI basic module and an FI finetuning module is developed to adjust the compensation quantity and prevent uncertain disturbance from undertrained ANN adaptively. Finally, the feasibility and efficiency of the proposed method are verified by the tracking experiments of step and square signals on the DC servo motor testbed. Experimental results show that the proposed FIS-enabled ANN feedforward compensation control method achieves lower overshoot, faster adjustment, and higher precision than other comparative control methods.

Machine learning based feedforward voltage control for buck converters in envelope tracking applications

Tracking the voltage envelope of 5G base station power supplies poses a significant long-term challenge. To achieve fast and accurate envelope tracking, an efficient and high-frequency power electronics circuit and an excellent voltage controller are essential. This paper presents a novel machine learning (ML) based feedforward control method for precise voltage reference tracking. The method is exemplified using a standard buck converter in both continuous conduction mode (CCM) and discontinuous conduction mode (DCM). A feedforward neural network (FNN) is trained to capture non-ideal and non-linear effects in real circuits. An automated data collection system, developed with MATLAB and Simulink, facilitates data acquisition and model training process. The trained FNN predicts the duty cycle for the buck converter to generate the targeted output voltage and output power. A simulation model is built to validate the effectiveness of the FNN controller in envelope tracking applications.

An Inexact Sequential Quadratic Programming Method for Learning and Control of Recurrent Neural Networks.

This article considers the two-stage approach to solving a partially observable Markov decision process (POMDP): the identification stage and the (optimal) control stage. We present an inexact sequential quadratic programming framework for recurrent neural network learning (iSQPRL) for solving the identification stage of the POMDP, in which the true system is approximated by a recurrent neural network (RNN) with dynamically consistent overshooting (DCRNN). We formulate the learning problem as a constrained optimization problem and study the quadratic programming (QP) subproblem with a convergence analysis under a restarted Krylov-subspace iterative scheme that implicitly exploits the structure of the associated Karush-Kuhn-Tucker (KKT) subsystem. In the control stage, where a feedforward neural network (FNN) controller is designed on top of the RNN model, we adapt a generalized Gauss-Newton (GGN) algorithm that exploits useful approximations to the curvature terms of the training data and selects its mini-batch step size using a known property of some regularization function. Simulation results are provided to demonstrate the effectiveness of our approach.

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.

Dynamic Output Feedback and Neural Network Control of a Non-Holonomic Mobile Robot.

This paper presents the design and synthesis of a dynamic output feedback neural network controller for a non-holonomic mobile robot. First, the dynamic model of a non-holonomic mobile robot is presented, in which these constraints are considered for the mathematical derivation of a feasible representation of this kind of robot. Then, two control strategies are provided based on kinematic control for this kind of robot. The first control strategy is based on driftless control; this means that considering that the velocity vector of the mobile robot is orthogonal to its restriction, a dynamic output feedback and neural network controller is designed so that the control action would be zero only when the velocity of the mobile robot is zero. The Lyapunov stability theorem is implemented in order to find a suitable control law. Then, another control strategy is designed for trajectory-tracking purposes, in which similar to the driftless controller, a kinematic control scheme is provided that is suitable to implement in more sophisticated hardware. In both control strategies, a dynamic control law is provided along with a feedforward neural network controller, so in this way, by the Lyapunov theory, the stability and convergence to the origin of the mobile robot position coordinates are ensured. Finally, two numerical experiments are presented in order to validate the theoretical results synthesized in this research study. Discussions and conclusions are provided in order to analyze the results found in this research study.

Feedforward neural network emulation of a PID continuous-time controller for quadcopter attitude digital control

<p>Quadcopters are popular UAVs owing to their compact size and maneuverability. Quadcopters are unmanned aircraft guided by remote control, and the demand for them is increasing due to their widespread surveillance, goods delivery, aerial photography, and defense applications. Nonlinear quadcopter operation makes control system implementation very challenging. In this paper, based on artificial intelligence (AI), we train a feedforward neural network (FFNN) controller of a traditional proportional integral derivative (PID). The conventional (PID) is generally tuned to improve the quadcopter control and performance. FFNN can perform offline learning between the inputs and outputs of the controller to learn its behavior. Once the learning is complete, we replace the PID controller with the neural network controller, to get a controller that can maintain system stability,and overcome the limitations of hardware implementation problems caused by the classical PID controller.</p>

Design and Experimental Characterization of Artificial Neural Network Controller for a Lower Limb Robotic Exoskeleton

This study aims to develop a lower limb robotic exoskeleton with the use of artificial neural networks for the purpose of rehabilitation. First, the PID control with iterative learning controller is used to test the proposed lower limb robotic exoskeleton robot (LLRER). Although the hip part using the flat brushless DC motors actuation has good tracking results, the knee part using the pneumatic actuated muscle (PAM) actuation cannot perform very well. Second, to compensate this nonlinearity of PAM actuation, the artificial neural network (ANN) feedforward control based on the inverse model trained in advance are used to compensate the nonlinearity of the PAM. Third, a particle swarm optimization (PSO) is used to optimize the PID parameters based on the ANN-feedforward architecture. The developed controller can complete the tracking of one gait cycle within 3.6 s for the knee joint. Among the three controllers, the controller of the ANN-feedforward with PID control (PSO tuned) performs the best, even when the LLRER is worn by the user and the tracking performance is still very good. The average Mean Absolute Error (MAE) of the left knee joint is 1.658 degrees and the average MAE of the right knee joint is 1.392 degrees. In the rehabilitation tests, the controller of ANN-feedforward with PID control is found to be suitable and its versatility for different walking gaits is verified during human tests. The establishment of its inverse model does not need to use complex mathematical formulas and parameters for modeling. Moreover, this study introduces the PSO to search for the optimal parameters of the PID. The architecture diagram and the control signal given by the ANN compensation with the PID control can reduce the error very well.

Learning for Precision Motion of an Interventional X-ray System: Add-on Physics-Guided Neural Network Feedforward Control

Tracking performance of physical-model-based feedforward control for interventional X-ray systems is limited by hard-to-model parasitic nonlinear dynamics, such as cable forces and nonlinear friction. In this paper, these nonlinear dynamics are compensated using a physics-guided neural network (PGNN), consisting of a physical model, embedding prior knowledge of the dynamics, in parallel with a neural network to learn hard-to-model dynamics. To ensure that the neural network learns only unmodelled Effects, the neural network output in the subspace spanned by the physical model is regularized via an orthogonal projection-based approach, resulting in complementary physical model and neural network contributions. The PGNN feedforward controller reduces the tracking error of an interventional X-ray system by a factor of 5 compared to an optimally tuned physical model, successfully compensating the unmodeled parasitic dynamics.

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.

Neural-Network-Based Nonlinear Model Predictive Control of Multiscale Crystallization Process

The purpose of this study was to develop an integrated control strategy for multiscale crystallization processes. An image analysis method using a deep learning neural network is used to measure the fine-scale information of the crystallization process, and the mathematical statistical method is adopted to obtain the mean size of the crystal population. A feedforward neural network is subsequently trained and employed in a nonlinear model predictive control formulation to obtain the optimal profile of the manipulated variable. The effectiveness of the proposed nonlinear model predictive control method is evaluated using alum cooling crystallization experiments. Experimental results demonstrate benefits of the proposed combination of feedforward neural network and nonlinear model predictive control method for the multiscale crystallization process.

Intelligent approach on sensorless control of permanent magnet synchronous generator

In this paper, a standalone permanent magnet synchronous generator (PMSG) system is designed to generate power at maximum power point (MPP). The variable speed operation of wind energy conversion system consists of PMSG, controlled rectifier and voltage source inverter co to the load. Proportional integral (PI), sliding mode (SM), and feed forward neural network (FFNN) control strategies are applied in field oriented control (FOC) at generator side converter. A comparative study on power generated at maximum power point (MPP) is done with these controllers using simulation. Hill climb search (HCS) method is applied to attain MPP. Load side inverter control strategy involves the PI and SM controllers in order to maintain the unity power factor and to control the active and reactive power for nonlinear load. The control strategies are modelled and simulated with MATLAB/Simulink. The effectiveness of proposed control method is demonstrated using simulation results.

Feedback-Error Learning for time-effective gait trajectory tracking in wearable exoskeletons.

The use of exoskeletons in gait rehabilitation implies user-oriented and efficient responses of exoskeletons' controllers with adaptability for human-robot interaction. This study investigates the performance of a bioinspired hybrid control, the Feedback-Error Learning (FEL) controller, to time-effectively track user-oriented gait trajectories and adapt the exoskeletons' response to dynamic changes due to the interaction with the user. It innovates with a controller benchmarking analysis. FEL combines a proportional-integral-derivative (PID) feedback controller with a three-layer neural network feedforward controller that learns the inverse dynamics of the exoskeleton based on real-time feedback commands. FEL validation involved able-bodied subjects walking with knee and ankle exoskeletons at different gait speeds while considering gait disturbances. Results showed that the FEL control accurately (tracking error <7%) and timely (delay <30 ms) tracked gait trajectories. The feedforward controller learned the inverse dynamics of the exoskeletons in a time compliant for clinical use and adapted to variations in the gait trajectories, such as speed and position range, while the feedback controller compensated for random disturbances. FEL was more accurate and time-effective controller for tracking gait trajectories than a PID control (error <27%, delay <260 ms) and a lookup table feedforward combined with PID control (error <17%, delay >160 ms). These findings aligned with FEL's time-effectiveness favors its use in wearable exoskeletons for repetitive gait training.

Statistical machine-learning-based predictive control using barrier functions for process operational safety

In this work, we present statistical model predictive control with Control Lyapunov-Barrier Functions (CLBF) built using machine learning approaches, and analyze closed-loop stability and safety properties in probability using statistical machine learning theory. A feedforward neural network (FNN) is used to construct the Control Barrier Function, and a generalization error bound can be obtained for this FNN via the Rademacher complexity method. The FNN Control Barrier Function is incorporated in a CLBF-based model predictive controller (MPC), which is used to control a nonlinear process subject to input constraints. The stability and safety properties of the closed-loop system under the sample-and-hold implementation of FNN-CLBF-MPC are evaluated in a statistical sense. We use a chemical process example to demonstrate the relation between various factors of building an FNN model and the generalization error, as well as the probabilities of closed-loop safety and stability for both bounded and unbounded unsafe sets.

Power Quality Improvement in Distribution System using SAPF with Feed Forward Neural Network and PQ theory

This article focuses on the Feed Forward Neural Network (FFNN) controller based on the Shunt Active Power Filter (SAPF) to reduce the harmonics of the distribution system. Electronic power loads are connected to distributed power plants using power converters, and these converters and electric power are a harmonic and active source of power that affects the operation of the power system network. The proposed method works mainly according to the capacitor power principle to maintain the DC switching power of the integrated filter and thus reduce the short response time in the event of a sudden difference in load. A complete model of the power system blocks of the proposed system was developed in MATLAB/Simulink. Keywords - Active Power Filter, Power Quality, Feed Forward Neural Network

Sampled-data stabilization analysis of neural-network-based control systems: A discontinuous bilateral looped-functional approach

This work is engaged in investigating the problem of stabilization of neural-network-based control systems (NNBCSs) via sampled-data (SD) control. Firstly, a time-dependent discontinuous bilateral looped-functional (DBLF) approach is proposed by fully considering the sampling state information, which relaxes the restriction of monotonicity. Then, a new lemma regarding sampled-data-based Lyapunov stability result is established for nonlinear systems, which is strictly proved by the Cauchy-Schwarz inequality. Moreover, the proposed lemma improves the conventional Lyapunov theorem and weakens the asymptotic stability conditions of continuous-time systems into discrete conditional expression. In addition, a SD three-layer fully connected feed-forward neural network (TLFCFFNN) controller is considered to reach the maximum interval of sampling. Finally, two numerical examples are presented to illustrate the effectiveness of the theoretical results.

Reduced Simulative Performance Analysis of Variable Step Size ANN Based MPPT Techniques for Partially Shaded Solar PV Systems

The rise in energy demand in the present scenario can be balanced with the help of solar Photovoltaic (PV) systems. But, the nonlinearity in I-V and P-V characteristics makes it very difficult to extract the maximum power of the solar PV. Also, the classical Maximum Power Point Tracking (MPPT) techniques fail to track the global Maximum Power Point (MPP) from the multiple local MPPs under Partial Shading Conditions (PSCs). In this work, a Variable Step Size ANN-based MPPT technique is proposed and it is compared with the other MPPT techniques in terms of steady-state behavior, settling time of converter power, power point tracing speed, oscillations of MPP, and operating efficiency. The compared MPPT techniques are Adaptive Perturb & Observe (AP&O), Adaptive Feed Forward Neural Network Controller (AFFNNC), Artificial Neural Network-based P&O (ANN-based P&O), ANN-based Incremental Conductance (ANN-based IC), ANN-based Hill Climb (ANN-based HC), and Radial Basis Functional Controller based Fuzzy (RBFC based Fuzzy). The boost converter is interfaced in the middle of the PV system and load to step-up the PV supply voltage. The performance of selected neural networks MPPT techniques is studied by utilizing a MATLAB/Simulink window.

Research on Structure Design and Control Method of Underwater Exploration Robot

For the sake of complete the detection of complex underwater environment in the engineering construction, this paper designs the structure of the underwater exploration robot according to the actual requirements of the project, and proposes a control method of the underwater exploration robot using neural network to improve the PID control parameters. Firstly, the overall structure and control system layout of underwater exploration robot are designed and planned according to the design requirements. Secondly, aiming at the problems of complex parameter setting and poor real-time performance of system parameters in traditional control methods, the automatic learning characteristics of feedforward neural network and PID controller are combined to realize the online alter of PID control unit parameters. Finally, according to the above method, a PID controller with automatic learning characteristics is designed to control the underwater exploration robot, which can achieve the optimal parameter combination of the underwater exploration robot's motion control, and improve its poor adaptability in underwater motion.

Feedforward Neural Network Controller for Quadrotor in the Presence of Payload and Wind Disturbances

Recently, Artificial Neural Networks (ANN) have been used in broad research fields such as control system engineering. Inspired by the human brain system, this algorithm uses the learning process to improve system performance using the training dataset. The ANN algorithm is based on a weighted connection between the neurons to deal with the controller design. These connection parameters represent one of the crucial elements necessary to apply the learning process efficiently. However, finding the connection parameters becomes problematic when the controlled system is complex, nonlinear, and subjected to external disturbances. This work uses the most available ANN architecture called the Feedforward Neural Network to stabilize the quadrotor system in reference trajectory tracking when subjected to model payload and Dryden turbulent wind. A Deterministic Algorithm (DA) trains the FNN weight and bias without initialization with another training method to improve the control performance. Besides, three simulation tests are performed to control the quadrotor system. The first simulation uses classical PID to collect the necessary training data for the learning process. Secondly, the DA algorithm trains the FNN parameters to design the controllers necessary to stabilize the quadrotor in-flight tests. The third simulation uses the most common training method: the Levenberg-Marquardt Backpropagation (LMB) algorithm to tune the FNN parameters. The final simulation tests these control methods' efficiency when the quadrotor system is subjected to the model payload and the Dryden turbulent wind. A comparison test between the proposed DA-based feedforward neural network, the classical PID, and the LMB-based feedforward neural network is performed in the nominal case (without disturbance) and in the case of adding disturbance. The test result is obtained using Matlab software and demonstrates much accuracy and efficacity of the proposed DA-based FNN controller compared to the other methods. This result considers the proposed control approach an additional innovative method-based neural network to stabilize the quadrotor's Unmanned Aerial Vehicles (UAV).

Adaptive feedforward RBF neural network control with the deterministic persistence of excitation

Adaptive feedforward RBF neural network control with the deterministic persistence of excitation

Power Optimization Control Scheme for Doubly Fed Induction Generator Used in Wind Turbine Generators

Scientists and researchers are exploring different methods of generating and delivering electrical energy in an economical and reliable way, enabling them to generate electricity focusing on renewable energy resources. All of these possess the natural property of self-changing behavior, so the connection of these separate independent controllable units to the grid leads to uncertainties. This creates an imbalance in active power and reactive power. In order to control the active and reactive power in wind turbine generators with adjustable speed, various control strategies are used to allay voltage and current variations. This research work is focused on the design and implementation of effective control strategies for doubly fed induction generator (DFIG) to control its active and reactive power. A DFIG system with its control strategies is simulated on MATLAB software. To augment the transient stability of DFIG, the simulation results for the active and reactive power of conventional controllers are compared with three types of feed forward neural network controllers, i.e., probabilistic feedforward neural network (PFFNN), multi-layer perceptron feedforward neural network (MLPFFN) and radial basic function feedforward neural network (RBFFN) for optimum performance. Conclusive outcomes clearly manifest the superior robustness of the RBFNN controller over other controllers in terms of rise time, settling time and overshoot value.