Reducing collision counts at intersections using reinforcement learning-based adaptive speed control

In this paper, the global adaptive speed tracking control algorithm with accurate parameters estimation of the surface mounted permanent magnet synchronous motors is presented. The adaptive controller is designed using second Lyapunov method and does not require over-parametrization. The algorithm may be used for fast self-commissioning of all motor parameters as well as adaptive speed controller with online identification of some parameters. During self-commissioning, in the first stage the stator resistance and inductance are estimated in motionless mode. Second stage provides identification of all mechanical parameters in speed control mode. Simulation results proof the effectiveness of the theoretical findings.

In this paper, event-triggered reinforcement learning-based adaptive tracking control is developed for the continuous-time nonlinear system with unknown dynamics and external disturbances. The critic and action neural networks are designed to approximate an unknown long-term performance index and controller, respectively. The dead-zone event-triggered condition is developed to reduce communication and computational costs. Rigorous theoretical analysis is provided to show that the closed-loop system can be stabilized. The weight errors and the filtered tracking error are all uniformly ultimately bounded. Finally, to demonstrate the developed controller, the simulation results are provided using an autonomous underwater vehicle model.

With the increase in tightening of carbon neutral policies, electric new energy vehicles have become a top priority in the development of the automotive industry. The integrated electric vehicle wheel (hub motor) can effectively optimize the vehicle structure and reduce the weight of the vehicle and reduce the energy consumption. At the same time, the feature that four tires can be driven separately is also conducive to the application of various new control algorithms. In this paper, a generalized adaptive PID control is experimented with the speed control of the hub motor, and a fuzzy control-based variable step method is proposed for the overshoot or even runaway problem caused by the inertia of the hub motor in the process of adaptive PID control. Experiments have demonstrated that this method can effectively reduce the amount of overshoot triggered by adaptive control and shorten the regulation time into a steady state.

Interior permanent magnet synchronous motors (IPMSMs) which are driven by matrix converters have many advantages, including one‐stage power conversion, high efficiency, bidirectional power transfer, near unity input power factor, and low input current harmonics. However, the virtual DC‐link voltages of matrix‐converters vary from 0.866 to and the current commutations of matrix‐converters create serious three‐phase stator harmonic currents as well. Moreover, matrix‐converter‐based IPMSM drive systems, in general, have very large external loads, which require advanced control methods. To solve these problems, two adaptive controllers, including a model reference adaptive controller and a self‐tuning controller, are investigated in this article for matrix‐converter PMSM drive systems. From the experimental results, the proposed two adaptive speed controllers provide excellent dynamic responses, which are superior to PI speed controllers, including faster transient responses, better tracking abilities, and greater robustness. As a result, this proposed matrix‐converter‐based IPMSM drive system is very suitable in industrial applications.

Real-time fault detection and active fault-tolerant control on vehicles can prevent potential accidents and ensure safe autonomous driving. Unlike existing methods which are designed for known faults or require perfect knowledge of vehicle models, this paper proposes a novel data-driven optimal adaptive fault-tolerant control method with model learning for automated vehicles to enable effective and safe driving under various faults. The actuator and sensor faults are integrated in one unified model to describe the vehicle dynamic behavior. The vehicle states are estimated online with only limited onboard sensors in vehicles available. An efficient fault detector using a modified sliding innovation sequence method is developed to ensure accurate fault detection. The dynamic model is adapted online using Gaussian Process based on vehicle running data, accurately representing the dynamic driving behavior. A novel learning-based adaptive fault-tolerant control method is designed to optimize the vehicle movement for autonomous path tracking under various faults with stochastic model predictive control. Simulation experiments validate the effectiveness of the proposed method in fault detection, model online learning, and accurate path tracking across various fault scenarios.

The paper deals with two concepts of a model reference adaptive control (MRAC) of the induction motor drive with elastic joint. The adaptive speed controller uses fuzzy neural network equipped with additional option for on-line tuning its chosen parameters. In the paper PI-type and sliding-mode fuzzy logic controllers are used as the speed controllers, whose connective weights are trained on-line according to the error between the state variable of the plant and the reference model. The torsional vibrations dumping abilities of the two-mass drive system with the field oriented control of induction motor are tested and compared for both analyzed neuro-fuzzy adaptive controllers. It is shown that torsional oscillations can be successfully suppressed in the proposed control structures, using only one basic feedback from the motor speed. The simulation results are verified in experimental tests, in the wide range of motor speed and parameters changes.

The switched reluctance motor drive (SRD) has obtained great attention as an AC stepless speed control system due to its large regulating scope, low cost and ruggedness. However, its strong nonlinearity and multivariable characteristic make it difficult to control. To solve the problem, this paper presents an approach of adaptive PWM speed control for switched reluctance motors (SRM) based on RBF neural network. This method builds up a speed controller based on RBF neural network which has powerful approximating ability and fast convergence property. The controller is trained off-line in advance, and then with the motor's operation, the on-line training of it makes its parameters vary with the environment in order to improve the control performance. In addition, another RBF network is constructed to offer gradient parameters, which is needed by the on-line training, via on-line identification. The results of experiments prove that the approach has lots of advantages in response speed, control accuracy and adaptability.

A Learning-based Adaptive Signal Control System with Function Approximation**This work is partially supported by J. Gust. Richert Foundation.

Due to the benefits of using hydrogen energy, such as enhanced system reliability, reduced CO<sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> emissions, and long term sustainability, the integration of hydrogen energy systems with power system sinmulti-energy microgrids (MEMGs) has attracted increasing research interest. The wide spread adoption of information and communication technologies has also increased the risk of cyber-attack son MEMGs. False data injection attacks (FDIAs), a common type of cyber attack, could falsify energy arbitrage commands, there by affecting these condary frequency regulation and degrading the power quality of the MEMG. These attacks not only affect the normal operations of the MEMG but also reduce its economic profit. To address these issues, this paper proposes a novel deep reinforcement learning-based adaptive resilient control scheme to coordinate the energy conversion between electricity and hydrogen to eliminate the frequency deviation and maximize daily total revenue. Unlike traditional optimization techniques, the propos d deep reinforcement learning-based approach enables real-time adaption to dynamic and adversarial conditions, allowing the system to learn optimal strategies and improve resilience against FDIAs. Case studies withreal-world data sets showthat as theattack injection level increasesfrom5%to20%, the proposed method maintains zero frequency deviation, with the total revenue decreasing by only 1.8%.

ABSTRACTThis research proposes a methodology based on Reinforcement Learning for deriving optimal adaptive control policies for CONWIP-type manufacturing systems. The Reinforcement Learning agent interacts with the production system and learns the best actions for each state. The goal is to minimize the total, average cost that consists of holding and backorder costs. The Reinforcement Learning-based approach is compared to the standard CONWIP, the Dynamic CONWIP and a special case of the Generalized Kanban system in a series of simulation experiments. The proposed approach is found to outperform all alternative control policies in all cases. The Generalized Kanban policy is reported to be a good approximation of the control policies that are derived by the Reinforcement Learning approach in situations where the demand rate is low. A numerical investigation of the total cost function properties for the CONWIP, Dynamic CONWIP and Generalized Kanban is also provided.

An adaptive model-following speed controller for induction motor drives is proposed and implemented. This adaptive controller is easy to implement since, for the controlled plant, only the output signal is fed back. In order to choose a reference model with the desired drive dynamic behavior, a two-degree-of-freedom (2-DOF) controller is designed, using a systematic procedure based on the prescribed drive specifications. The reference model of the adaptive system is chosen as the transfer function of the controlled drive system by the 2-DOF controller in the nominal case. An adaptation signal is augmented to let the prescribed specifications be maintained under a wide range of operating conditions.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Learning-based adaptive control and analysis for nonlinear systems with state-dependent constraint

In the paper, a nonlinear adaptive speed controller for a permanent magnet synchronous motor based on a newly developed adaptive backstepping approach (PMSM) is presented. The exact input-output feedback linearisation control law is first introduced without any uncertainties in the system. However, in real applications, the parameter uncertainties such as the stator resistance and viscous constant and load torque disturbance have to be considered. In this case, the exact input-output feedback linearisation approach is not very effective, because it is based on the exact cancellation of the nonlinearity. To compensate the uncertainties and the load torque disturbance, the input-output feedback linearisation approach is first used to compensate the nonlinearities in the nominal system. Then, a nonlinear adaptive backstepping control law and parameter uncertainty, and load torque disturbance adaptation laws, are derived systematically by using adaptive backstepping technique. Simulation and experimental results clearly show that the proposed adaptive scheme can track the speed reference signal generated by a reference model, successfully, and that the scheme is robust to the parameter uncertainties and load torque disturbance.

Chapter 9 - Data-driven methods for flexible distribution networks operation

This paper presents the application of adaptive algorithm for the speed control of Switched Reluctance Motor (SRM) with load series DC machine. This system is feeding by Photovoltaic (PV) system. SRM is a highly nonlinear control plant and operates in saturation to maximize the torque output. A systematic approach to the modeling of highly nonlinear SRM drive system which includes the motor, converter, and its electronic controller is presented. Hysteresis current controlled mid-point converter is used to feed four phase, 8/6 pole SRM. Nonlinearity caused by magnetic saturation is accounted for accurate and real-time simulation of drive performance by considering experimental data of magnetization and static torque characteristics. Performance analysis of SRM drive is reported for a wide range of operating conditions via starting, reversal and load perturbation dynamics the performance indices of SRM drive system operating with adaptive controller for speed control and using the PI classical controller turn on angle for starting operation (dynamic operation). The results illustrate that the approach used are suitable for dynamic analysis and steady state of SRM fed from a PV system. It ensures smooth starting without initial hesitation and reduce the torque ripples. This proposed is explain the comparison between two technique for SRM (four phases) control system. First controller is adaptive speed control and the second controller is adaptive torque control.