A fast on-line neural-network training algorithm for a rectifier regulator

Abstract

Abstract-This paper addresses the problem of deadbeat control in fully controlled high-power-factor rectifiers. Improved deadbeat control can be achieved through the use of neuralnetwork-based predictors for the input-current reference to the rectifier. In this application, on-line training is absolutely required. In order to achieve sufficiently fast on-line training, a new random-search algorithm is presented and evaluated. Simulation results show that this type of network training yields equivalent performance to standard backpropagation training. Unlike backpropagation, however, the random weight change (RWC) method can be implemented in mixed digital/analog hardware for this application. The paper proposes a very large-scale integration (VLSI) implementation which achieves a training epoch as low as 8 s.Index Terms-Current regulator, neural network, on-line training, power electronic rectifier.

Lyapunov‐based training algorithm applied to a continually on line‐trained ANN used in the current‐loop control of a single‐phase switched rectifier

“…A typical solution to this is the use of intelligent controllers which are able to adjust to its present operating conditions. The use of artificial neural networks (ANNs) has previously been proposed to obtain an estimated value of the system current used to feed a conventional current controller with a state predictor [4]. Furthermore, with the use of intelligent controllers the converter can be used to feed a completely new system without the need to recalibrate or expensive commissioning of the controller.…”

Integration of grey model and neural network for robotic application

“…As opposed to batch training, on-line training adjusts the network according to a single training pattern at a time. It is particularly well suited to the simultaneous execution of signal prediction and learning to improve performance, due to the use of patternby-pattern weight updating [9], [10]. In [11], the author found that if the learning rate is low enough, a weighted search will converge.…”

Simplified Control Structure for Current Control of Single Phase Rectifiers Using COT-ANN-PWM

“…PWM based systems make use of the traditional PI controller, embedded in the control software, for the current loop [2], but precise tuning is needed for optimum operation of the controller, and in general this can be achieved only for a particular operating point. The use of artificial neural networks has previously been proposed to obtain an estimated value of the system current used to feed a conventional current controller with a state predictor in a three phase rectifier [3]. A direct current controller has also been proposed [4], but its capabilities were evaluated only by simulations.…”